Insurance business real-time risk early warning method and system under dynamic risk assessment

Abstract

The invention provides an insurance service real-time risk early warning method and system under dynamic risk assessment. The method comprises the steps of obtaining multi-modal risk data including time, space and industry dimensions in an insurance service; separating a core risk element region from the multi-modal risk data through a mask mechanism, and generating an initial risk feature map with a mask mark; inputting the initial risk feature map into a dynamic risk assessment network of an encoder-decoder architecture; wherein both the encoder and the decoder comprise multi-stage cascaded risk processing modules, and each risk processing module is integrated with a gating loop unit and is configured with a binary risk gate for dynamically controlling an update path of risk characteristics; and generating a real-time early warning signal by combining a preset risk level threshold based on the target risk characteristics output by the decoder. According to the invention, the processing path of the risk features can be dynamically adjusted, and the timeliness and accuracy of risk early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a real-time risk warning method and system for insurance business under dynamic risk assessment. Background Art

[0002] In modern insurance, risk assessment and early warning are critical components for ensuring the industry's robust operations. With the continuous expansion of insurance business and the increasing complexity of business scenarios, traditional risk assessment methods have gradually exposed numerous issues. Currently, most insurance risk assessments employ a static model, relying solely on historical data and a fixed risk indicator system for risk assessment.

[0003] On the one hand, it's difficult to effectively process the dynamic, multi-dimensional risk data in the insurance business, encompassing time, space, and industry. For example, information such as the timing and geographical distribution of natural disasters, and the timing and coverage of industry policy adjustments, is difficult to integrate and deeply analyze using traditional assessment models, resulting in insufficiently comprehensive risk assessments. On the other hand, the traditional risk assessment network has a simple structure and mostly uses fixed feature extraction and processing processes. It is difficult to adjust the processing path of risk features in real time according to different business scenarios and risk changes, resulting in poor timeliness and accuracy of risk warnings, which cannot meet the urgent needs of real-time risk management in the insurance business. Summary of the Invention

[0004] The main purpose of the present invention is to provide a real-time risk warning method and system for insurance business under dynamic risk assessment, aiming to overcome the shortcomings of current risk warning methods in terms of timeliness and accuracy.

[0005] To achieve the above objectives, the present invention provides a real-time risk warning method for insurance business under dynamic risk assessment, comprising the following steps: Acquire multimodal risk data in insurance business, including time, space, and industry dimensions; The core risk element area is separated from the multimodal risk data through a masking mechanism to generate an initial risk feature map with mask marks; The initial risk profile is fed into a dynamic risk assessment network with an encoder-decoder architecture. Both the encoder and decoder contain multi-stage cascaded risk processing modules, each of which integrates a gated recurrent unit and is configured with a binary risk gate to dynamically control the update path of the risk profile. Based on the target risk characteristics output by the decoder and combined with the preset risk level threshold, a real-time warning signal is generated.

[0006] Furthermore, the encoder is used to perform time dimension hybrid mapping on the initial risk feature map, including: After normalizing the initial risk feature map, the target update probability of the current window is calculated based on the state update probability and change value of the previous window through the multi-stage time window iteration mechanism of the gated recurrent unit; The target update probability is rounded to generate the gating parameters of the binary risk gate. The gate parameters are used to determine whether to update the hidden state features or keep the original values. After full connection processing, the temporal risk features are output; The encoder is also used for spatial dimension mixed mapping, including: Split the time series risk characteristics into regional risk sub-characteristics and industry risk sub-characteristics along the risk dimension channel; Perform convolution operations on the regional risk sub-features along the time axis in the sub-dimensions to generate regional risk propagation features; The regional transmission characteristics and industry risk sub-characteristics are integrated through element-by-element product to obtain the risk characteristic matrix that represents the spatiotemporal correlation. The decoder performs multi-scale feature reconstruction on the risk feature matrix representing spatiotemporal correlation to obtain the target risk feature.

[0007] Furthermore, the training process of the dynamic risk assessment network includes: Construct training samples; Extract features from training samples through a dynamic risk assessment network to generate risk feature prediction distribution; The variance matrix of the predicted distribution is calculated, and the target loss function is constructed by combining the difference between the predicted distribution and the risk label and the variance matrix to perform end-to-end training on the dynamic risk assessment network.

[0008] Furthermore, constructing the target loss function includes: In the first stage, the accuracy of the model on risk variance is constrained based on the differences in spatiotemporal coordinates and variance matrix between the predicted distribution and the risk label; In the second stage, weights are assigned to features of different dimensions based on risk variance to strengthen feature learning in high uncertainty areas; The dynamic risk assessment network is trained sequentially based on two-stage loss functions.

[0009] Furthermore, the variance matrix includes time dimension variance, space dimension variance, and time-space coupling variance. The accuracy of the constraint model on risk variance includes: Calculate the characteristic differences between the predicted distribution and the risk label in time, space, and spatiotemporal coupling dimensions; The first risk loss term is obtained by weighting the feature differences with a function whose exponent is the variance of each dimension; Determine the second risk loss item based on the risk level distribution difference and variance matrix; The objective loss function of the first stage is obtained by weighted summing of the first risk loss term and the second risk loss term.

[0010] Furthermore, the dynamic control mechanism of the binary risk gate includes: For the current time window, the target update probability matrix is ​​predicted through the logistic regression model based on the state update probability sequence and change gradient of the previous window; The target update probability matrix is ​​binarized to generate a gating mask, and the attention head of the gated recurrent unit is selectively activated according to the gating mask to achieve adaptive weighted aggregation of temporal risk features.

[0011] Furthermore, the multimodal risk data includes policyholder behavior sequences, geographical area risk indicators, and industry claim rates.

[0012] The present invention also provides a real-time risk warning system for insurance business under dynamic risk assessment, comprising: The acquisition module is used to obtain multimodal risk data in insurance business, including time, space, and industry dimensions; A generation module is used to separate the core risk element area from the multimodal risk data through a mask mechanism and generate an initial risk feature map with mask labels; An input module, which feeds the initial risk profile into a dynamic risk assessment network based on an encoder-decoder architecture. Both the encoder and decoder contain multi-stage cascaded risk processing modules, each of which integrates a gated recurrent unit and a binary risk gate to dynamically control the update path of the risk profile. The early warning module is used to generate a real-time early warning signal based on the target risk characteristics output by the decoder and the preset risk level threshold.

[0013] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0015] The present invention provides a real-time risk warning method and system for insurance business under dynamic risk assessment, including: obtaining multimodal risk data containing time, space, and industry dimensions in the insurance business; separating the core risk element area from the multimodal risk data through a masking mechanism to generate an initial risk feature map with mask markings; inputting the initial risk feature map into a dynamic risk assessment network of an encoder-decoder architecture; wherein the encoder and decoder both include multi-stage cascaded risk processing modules, each risk processing module integrates a gated loop unit and is configured with a binary risk gate for dynamically controlling the update path of the risk feature; based on the target risk feature output by the decoder, a real-time warning signal is generated in combination with a preset risk level threshold. In the present invention, multimodal risk data containing time, space, and industry dimensions in the insurance business is obtained, and real-time risk warnings for the insurance business are performed from multiple dimensions; at the same time, through the special network structure of the dynamic risk assessment network, the processing path of the risk feature can be dynamically adjusted, thereby improving the timeliness and accuracy of the risk warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the steps of a method for real-time risk warning of insurance business under dynamic risk assessment in one embodiment of the present invention; Figure 2 This is a structural block diagram of a real-time risk warning system for insurance business under dynamic risk assessment in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0017] The implementation, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] Reference Figure 1 In one embodiment of the present invention, a method for real-time risk warning of insurance business under dynamic risk assessment is provided, comprising the following steps: Step S1, obtaining multimodal risk data in insurance business including time, space and industry dimensions; Step S2: Separate the core risk element area from the multimodal risk data through a masking mechanism to generate an initial risk feature map with mask markings; Step S3: Input the initial risk profile into a dynamic risk assessment network with an encoder-decoder architecture; both the encoder and decoder include multi-stage cascaded risk processing modules, each of which integrates a gated recurrent unit and is configured with a binary risk gate to dynamically control the update path of the risk profile; Step S4: Generate a real-time warning signal based on the target risk characteristics output by the decoder and a preset risk level threshold.

[0020] In this embodiment, as described in step S1 above, during the operation of the insurance business, the risk situation is affected by a variety of factors, which are presented in the form of data of different dimensions and modalities. The data of the time dimension includes the time series of historical risk events, the risk change trend of each stage in the business cycle, etc.; the data of the spatial dimension covers the geographical location information of the risk events, the differences in risk characteristics of different regions, etc.; the data of the industry dimension includes changes in industry policies, industry development trends, risk cases in the same industry, etc. Through multi-source data acquisition technology, such as from insurance business databases, external industry data platforms, geographic information systems and other channels, these data of different dimensions and modalities are collected and integrated to form a comprehensive and information-rich multimodal risk data set, which provides basic data support for subsequent risk assessment and early warning.

[0021] As described in step S2 above, the core risk element areas are separated from the multimodal risk data through a masking mechanism to generate an initial risk feature map with mask markings. Due to the large amount of redundant information in the multimodal risk data, not all data are equally important for risk assessment. The masking mechanism filters and analyzes the multimodal risk data by setting preset rules and algorithms. Its principle is similar to adding a "screening filter" to the data, which can identify the core risk element areas in the data that play a key role in risk assessment, such as high-risk areas in a specific time period, and business sectors that are significantly affected by industry policies. These core risk element areas are marked and combined with the original data to be converted into a visual initial risk feature map. The feature map highlights the core risk elements through mask markings, so that subsequent processing can focus on key risk information and improve the efficiency and accuracy of risk assessment.

[0022] As described in step S3 above, the initial risk profile is input into a dynamic risk assessment network based on an encoder-decoder architecture. Both the encoder and decoder contain multi-stage cascaded risk processing modules, each of which integrates a gated recurrent unit (GRU) and is configured with a binary risk gate to dynamically control the update path of the risk profile. The initial risk profile, as visual data containing core risk elements, first enters the encoder. The multi-stage cascaded risk processing modules in the encoder perform step-by-step feature extraction and abstraction on the input feature profile. The gated recurrent unit (GRU) in each risk processing module is a recurrent neural network unit that captures the dynamic changes in risk profiles over time or other dimensions. It uses a gating mechanism to control the inflow and outflow of information, selectively retaining and updating risk profiles. For example, when processing risk profiles related to time series, the GRU can memorize past risk states and, based on current input, determine how to update the current risk profile representation.

[0023] The configured binary risk gate further enhances the flexibility and dynamism of risk signature processing. The binary risk gate outputs either 0 or 1. When the risk gate output is 1, the current risk processing module is allowed to update and transform the input risk signature; when the output is 0, the current module is blocked from updating the risk signature, allowing the risk signature to be directly passed to the next-level module. This mechanism dynamically selects the risk signature update path based on different business scenarios and risk changes, avoiding unnecessary signature transformations while preserving key risk information.

[0024] After processing by the encoder, the risk signature is abstracted into a higher-level feature representation and then passed to the decoder. The decoder, similarly composed of a multi-level cascade of risk processing modules, decodes and reconstructs the abstract features output by the encoder, restoring them into target risk signatures that more closely resemble the original data and contain deeper risk information. During the decoder's processing, the GRU and binary risk gates in the risk processing modules continue to optimize and adjust the risk signatures, ensuring that the final target risk signature accurately reflects the current risk profile of the insurance business.

[0025] As described in step S4 above, a real-time warning signal is generated based on the target risk characteristics output by the decoder and in combination with a preset risk level threshold. The preset risk level threshold is pre-set based on multiple factors such as historical risk data of the insurance business, industry standards, and the company's own risk tolerance, and is used to divide the boundary values ​​of different risk levels. After the decoder outputs the target risk characteristics, it is compared and analyzed with the preset risk level threshold. If the risk level reflected by the target risk characteristics exceeds a certain level threshold, a corresponding real-time warning signal is generated according to the corresponding risk level. The warning signal can promptly notify relevant personnel of the insurance business so that they can take targeted risk response measures to achieve real-time monitoring and effective management of insurance business risks.

[0026] In the present invention, multimodal risk data including time, space and industry dimensions of insurance business is obtained, and real-time risk warning of insurance business is carried out from multiple dimensions; at the same time, through the special network structure of the dynamic risk assessment network, the processing path of risk characteristics can be dynamically adjusted, thereby improving the timeliness and accuracy of risk warning.

[0027] In one embodiment, the encoder is used to perform time-dimensional hybrid mapping on the initial risk profile, including: After normalizing the initial risk feature map, the target update probability of the current window is calculated based on the state update probability and change value of the previous window through the multi-stage time window iteration mechanism of the gated recurrent unit; The target update probability is rounded to generate the gating parameters of the binary risk gate. The gate parameters are used to determine whether to update the hidden state features or keep the original values. After full connection processing, the temporal risk features are output; The encoder is also used for spatial dimension mixed mapping, including: Split the time series risk characteristics into regional risk sub-characteristics and industry risk sub-characteristics along the risk dimension channel; Perform convolution operations on the regional risk sub-features along the time axis in the sub-dimensions to generate regional risk propagation features; The regional transmission characteristics and industry risk sub-characteristics are integrated through element-by-element product to obtain the risk characteristic matrix that represents the spatiotemporal correlation. The decoder performs multi-scale feature reconstruction on the risk feature matrix representing spatiotemporal correlation to obtain the target risk feature.

[0028] In this embodiment, during the time dimension hybrid mapping process of the encoder: First, the initial risk profile is normalized. This step ensures consistency in the format and scope of the data in the initial risk profile, eliminates interference from data differences on subsequent analysis, and lays the foundation for accurate risk profile extraction.

[0029] Next, a multi-stage time window iteration mechanism using a gated recurrent unit is employed. This mechanism divides time into multiple windows and, when calculating the target update probability for the current window, considers the state update probability and change value from the previous window. This is similar to planning the direction of current work based on past work achievements and trends. This approach captures the dynamic evolution of risk characteristics over time and determines the degree to which the current window retains historical information and incorporates new information.

[0030] The calculated target update probability is then rounded to an integer to generate the gating parameter for the binary risk gate. The binary risk gate is like an intelligent switch, with a gating parameter that has only two states: 0 and 1. When the gating parameter is 1, the switch is on, allowing the current risk processing module to update the hidden state features and incorporate new risk information. When the gating parameter is 0, the switch is off, and the hidden state features are directly passed to the next-level module, maintaining their original state.

[0031] Finally, after full connection processing, the updated latent state features are integrated and transformed to output a temporal risk feature that contains temporal dependencies. The resulting temporal risk feature contains information about how risks change over time, providing a time-series risk basis for subsequent analysis.

[0032] In the encoder's spatial dimension hybrid mapping process: The first step is to segment the obtained time-series risk characteristics into regional risk sub-characteristics and industry risk sub-characteristics according to the risk dimension channels. This is equivalent to breaking down a complex risk entity into regional risk components related to geographic space and industry risk components related to industry attributes, facilitating in-depth analysis of different types of risks.

[0033] The second step is to perform convolution operations on the sub-dimensions along the time axis for the regional risk sub-features. The convolution operation acts like an "information detector." By sliding along the time dimension, it can capture the changing patterns of regional risk spread at different points in time, exploring the diffusion trends and evolution patterns of regional risks over time, and thus generate regional risk propagation characteristics.

[0034] The third step is to integrate the regional risk propagation signature with the industry risk sub-signatures through element-by-element multiplication. This operation multiplies the elements of the two signatures at corresponding locations, combining the temporal evolution of regional risk with industry risk information to produce a risk signature matrix representing spatiotemporal correlations. Each element in this matrix reflects the interaction between regional and industry risk at a specific spatiotemporal location, providing a comprehensive mapping of risk across both spatial and temporal dimensions.

[0035] When the decoder reconstructs multi-scale features of the risk feature matrix representing spatiotemporal correlations, it relies on its multi-level cascade risk processing module architecture, combined with gated recurrent units and binary risk gates, to achieve dynamic processing of risk features. The specific process is as follows: First, the risk feature matrix representing spatiotemporal correlations is input into the decoder and processed by the first-level risk processing module. The gated recurrent unit within the module analyzes the current input risk features and, combined with historical risk status information, selectively retains or updates them through a gating mechanism, capturing their dynamic temporal evolution. Simultaneously, a binary risk gate calculates and outputs a state value of 0 or 1 based on a preset algorithm and input features. When the value is 1, the current module initiates the update of the risk features, allowing the module to transform and extract the input features. When the value is 0, the update is blocked, allowing the risk features to be directly passed to the next-level module, thereby preliminarily screening and controlling the processing path of risk features. After the first level of processing, the risk signatures enter the next-level risk processing module, repeating the collaborative processing process of the gated recurrent unit and the binary risk gate. Different cascaded risk processing modules, configured with different parameters, mine risk signatures at multiple scales. For example, some modules focus on extracting local, detailed risk features, while others focus on capturing global, macroscopic risk characteristics. At each level of processing, the gated recurrent unit continuously updates the time dimension of the risk signatures, while the binary risk gate dynamically adjusts the processing path based on changes in the risk signatures, ensuring that key risk signatures at different scales are retained and enhanced. After risk signatures are processed sequentially by multiple risk processing modules, a feature set containing multi-scale information is formed. The decoder then integrates the risk signatures generated by each level of processing through a specific fusion mechanism. During the fusion process, the gated recurrent unit further refines the fused signatures, optimizing the inter-feature correlations. The binary risk gate then performs a final update path control based on the overall fused risk signature state, removing redundant information and strengthening the representation of core risk signatures. Finally, the fused risk signature undergoes post-processing, including standardization, to produce the final target risk signature. This target risk signature not only fully preserves the temporal and spatial correlation characteristics of risk but also incorporates multi-scale feature representations and temporal evolution information, providing a high-quality data foundation for subsequent accurate risk level assessment and real-time early warning based on risk signatures.

[0036] In one embodiment, the training process of the dynamic risk assessment network includes: Construct training samples; Extract features from training samples through a dynamic risk assessment network to generate risk feature prediction distribution; The variance matrix of the predicted distribution is calculated, and the target loss function is constructed by combining the difference between the predicted distribution and the risk label and the variance matrix to perform end-to-end training on the dynamic risk assessment network.

[0037] In this embodiment, it is first necessary to construct a training sample set containing multimodal risk data and corresponding risk labels. The module collects original risk data in the time, space, and industry dimensions of the insurance business, such as historical claims records, geographic area risk indexes, industry policy changes, etc. The original data is preprocessed, including data cleaning to remove noise and outliers, feature standardization to unify the dimensions, and time series alignment to ensure the consistency of the data time dimension. Then, according to factors such as the severity and scope of impact of historical risk events, each sample is labeled with a corresponding risk level label to form a training sample pair containing input features and target labels. In order to enhance the generalization ability of the model, the training samples will also be subjected to data enhancement processing, such as time series perturbation, spatial area masking, small feature value offset and other operations to expand the diversity of the training samples.

[0038] The constructed training samples are then input into the dynamic risk assessment network for feature extraction. The network's encoder first performs a spatiotemporal hybrid mapping of the multimodal risk data. This involves extracting temporal risk features through a multi-stage time window iteration mechanism of gated recurrent units, combined with a binary risk gate to dynamically control the feature update path. Simultaneously, features are segmented along the risk dimension, processing regional and industry risks separately and fusing them together to generate a risk feature matrix representing spatiotemporal correlations. The decoder reconstructs multi-scale features from this matrix, similarly utilizing a multi-stage cascaded risk processing module and binary risk gates to mine risk features at different scales, ultimately outputting the target risk features.

[0039] Unlike traditional deterministic outputs, this network uses probabilistic modeling to output a predicted distribution of risk characteristics. Specifically, the network predicts not only the mean of the risk characteristic but also its variance, forming a complete probability distribution that reflects the uncertainty of the model's prediction of the risk characteristic.

[0040] Finally, the variance matrix of the predicted distribution is calculated. This matrix contains the covariance information between the various dimensions of the risk profile and reflects the structure of predictive uncertainty. The objective loss function is constructed by combining the difference between the predicted distribution and the risk label, along with the variance matrix. This loss function is designed as the sum of two parts: the first is the negative log-likelihood loss between the predicted distribution and the true risk label, which measures the degree of deviation between the center of the predicted distribution and the true label, encouraging the model to accurately predict the risk level. The second part is a regularization term based on the variance matrix, which penalizes the uncertainty of the model's predictions. By minimizing this part of the loss, the model learns to reduce variance in high-confidence regions and increase variance in low-confidence regions, thereby better capturing risk uncertainty. In particular, the variance matrix plays a dual role in the loss function: first, it uses covariance information to adjust the weights of different risk dimensions, allowing the model to focus more on key risk factors; second, it uses variance information to quantify the model's confidence in its predictions, effectively modeling risk uncertainty.

[0041] In this example, an end-to-end training approach is employed, using a backpropagation algorithm to simultaneously optimize the encoder and decoder parameters to minimize the target loss function. During training, the learning rate is dynamically adjusted to ensure convergence stability, and an early stopping strategy is employed to prevent overfitting. The resulting model not only accurately predicts risk levels but also quantifies the uncertainty of these predictions.

[0042] In one embodiment, constructing the target loss function includes: In the first stage, the accuracy of the model on risk variance is constrained based on the differences in spatiotemporal coordinates and variance matrix between the predicted distribution and the risk label; In the second stage, weights are assigned to features of different dimensions based on risk variance to strengthen feature learning in high uncertainty areas; The dynamic risk assessment network is trained sequentially based on two-stage loss functions.

[0043] In this embodiment, in the first stage, at the initial stage of model training, the focus is on establishing an accurate constraint mechanism for risk prediction variance. First, the difference between the predicted distribution and the true risk label in spatiotemporal coordinates is calculated. This difference reflects the degree of deviation in the model prediction. At the same time, the variance matrix of the predicted distribution is extracted, which contains uncertainty information for each risk feature dimension. By designing a specific loss function term, the spatiotemporal coordinate difference is combined with the variance matrix to form a constraint on the risk variance. Specifically, the loss function penalizes the deviation between the predicted variance and the actual uncertainty, prompting the model to learn a variance representation that accurately reflects risk fluctuations. For example, for spatiotemporal regions where risk changes drastically, the model should output a larger variance value; while for stable regions, the variance value should be reduced accordingly. This constraint mechanism ensures that the model can not only predict the average level of risk, but also accurately quantify the uncertainty of the prediction, providing a reliable basis for the allocation of feature weights in subsequent stages.

[0044] In the second phase, after the model has initially mastered an accurate representation of risk variance, it enters the dynamic feature weighting phase. Based on the risk variance learned in the first phase, adaptive weights are assigned to features of different dimensions. Specifically, for high-variance regions (i.e., areas with greater uncertainty in the model's predictions), the weights of corresponding feature dimensions are increased, allowing the model to devote more learning resources to these areas and strengthen its ability to capture complex risk patterns. For low-variance regions (i.e., areas with more certainty in the model's predictions), feature weights are appropriately reduced to avoid overfitting. This risk variance-based weighting mechanism is essentially a dynamic attention mechanism, enabling the model to focus on high-risk, high-uncertainty regions and features, enhancing its sensitivity to key risk factors. In this way, the model can more effectively learn feature representations for different risk scenarios and enhance its ability to identify abnormal risks.

[0045] A two-stage sequential training approach is adopted to fully leverage the advantages of different loss functions. In the first stage, the focus is on optimizing the risk variance constraint loss function to enable the model to establish accurate risk uncertainty representation capabilities. This stage of training provides a reliable variance estimation basis for the subsequent feature weight allocation. In the second stage, the variance estimation parameters obtained from the first stage of training are fixed, and the feature weight allocation loss function based on risk variance is optimized instead. Through this staged training, the model gradually transitions from learning the basic representation of risk to fine-grained modeling of high-uncertainty areas, achieving hierarchical learning of risk features. This strategy effectively solves the convergence difficulties caused by the simultaneous optimization of multiple objectives in traditional training methods, enabling the model to more stably learn a comprehensive and in-depth risk feature representation, ultimately improving the warning accuracy and reliability of the dynamic risk assessment network.

[0046] In one embodiment, the variance matrix includes time dimension variance, space dimension variance, and time-space coupling variance. The accuracy of the constraint model on risk variance includes: Calculate the characteristic differences between the predicted distribution and the risk label in time, space, and spatiotemporal coupling dimensions; The first risk loss term is obtained by weighting the feature differences with a function whose exponent is the variance of each dimension; Determine the second risk loss item based on the risk level distribution difference and variance matrix; The objective loss function of the first stage is obtained by weighted summing of the first risk loss term and the second risk loss term.

[0047] In this embodiment, first, the characteristic differences between the predicted distribution and the real risk label in time, space, and time-space coupling dimensions are calculated respectively. In the time dimension, the trend of predicted risk changes over time is compared with the difference in the real risk in the corresponding time series, for example, the degree of consistency between the high-risk period predicted by the model and the actual period of occurrence is evaluated; the spatial dimension focuses on the deviation between the distribution of predicted risks in geographical areas and the spatial distribution of real risks, such as checking the accuracy of the model's prediction of risk levels in different regions; the time-space coupling dimension focuses on the risk differences under the joint action of time and space factors, and analyzes whether the model accurately captures the comprehensive risk characteristics of specific time points and specific regions. By quantifying the differences in these dimensions, the degree of deviation between the predicted results and the actual situation in multiple dimensions is obtained, providing basic data for subsequent weighted processing.

[0048] Next, using the variances of the time, space, and spatiotemporal coupling dimensions in the variance matrix, a function exponentially based on the variance of each dimension is constructed to weight the feature differences calculated in step 1. Variance reflects the uncertainty of risk in the corresponding dimension; a larger variance indicates greater volatility and uncertainty in the risk of that dimension. Therefore, a function exponentially based on variance gives greater weight to feature differences in high-variance dimensions during the weighting process, making the model more sensitive to prediction bias in high-uncertainty dimensions during training. The weighted feature differences in the time, space, and spatiotemporal coupling dimensions are summed to obtain the first risk loss term. This process enables the model to dynamically adjust its focus on feature differences in different dimensions during training based on the degree of risk uncertainty, thereby enhancing learning in areas of high risk and high uncertainty.

[0049] Furthermore, in addition to considering the feature differences in each dimension, the model needs to be further constrained from the perspective of risk level distribution. Analyze the difference between the risk level distribution of the predicted distribution and the risk level distribution of the actual risk labels. For example, compare the proportion of low, medium, and high risk level samples predicted by the model to see if it is consistent with the actual situation. At the same time, combined with the variance matrix, consider the uncertainty of the risk level distribution differences in different time and space dimensions. Through a specific calculation method, the risk level distribution differences are combined with the uncertainty reflected by the variance matrix to determine the second risk loss term. The second risk loss term ensures that the model is not only close to the true value in specific feature dimensions, but also accurately reflects the actual situation from the perspective of the overall risk level distribution, avoiding systematic risk level misjudgments in the model.

[0050] Finally, the first and second risk loss terms are summed according to pre-set weights to form the target loss function for the first stage. The weightings comprehensively consider the importance of different loss terms in constraining the accuracy of the model's risk variance and can be adjusted through experimentation or based on domain knowledge. By minimizing this target loss function, the dynamic risk assessment network is driven to optimize its parameters. During training, the model continuously corrects for multi-dimensional discrepancies between the predicted distribution and the true risk labels, improving its ability to accurately predict risk variance.

[0051] In one embodiment, the dynamic control mechanism of the binary risk gate includes: For the current time window, the target update probability matrix is ​​predicted through the logistic regression model based on the state update probability sequence and change gradient of the previous window; The target update probability matrix is ​​binarized to generate a gating mask, and the attention head of the gated recurrent unit is selectively activated according to the gating mask to achieve adaptive weighted aggregation of temporal risk features.

[0052] In this embodiment, when processing risk characteristics in the current time window, the binary risk gate first uses the state update probability sequence and change gradient of the previous window as input to build a prediction foundation. The state update probability sequence records the update probability of each risk characteristic dimension in the previous window, reflecting the model's confidence in the historical risk state. The change gradient describes the rate of change of these probabilities over time, capturing the dynamic evolution of risk characteristics.

[0053] These two sets of data are processed using a logistic regression model. Based on the input features, the model learns the underlying patterns of risk state changes and outputs a target update probability matrix. Each element in this matrix represents the probability of updating the corresponding risk feature dimension in the current window, with values ​​ranging from [0, 1]. This design enables the model to dynamically predict the risk feature dimensions that require current attention based on historical risk evolution patterns, enabling proactive responses to risk changes.

[0054] The target update probability matrix is ​​then converted to a gated mask through a threshold binarization operation. A threshold of 0.5 is typically used, with elements above this value converted to 1 and elements below this value converted to 0. The resulting gated mask is a binary matrix where a value of 1 indicates that the corresponding risk feature dimension should be given priority attention in the current window, while a value of 0 indicates that it can be temporarily ignored.

[0055] In the gated recurrent unit, each attention head is responsible for capturing different types of risk feature correlation patterns. The gated mask selectively activates these attention heads, enabling adaptive weighted aggregation of temporal risk features. Specifically, when a position in the gated mask is 1, the corresponding attention head is activated, and the weight of the risk feature output by it increases, indicating that this feature dimension is more important for the current risk assessment. Conversely, when the gated mask is 0, the output weight of the corresponding attention head is suppressed, reducing the interference of irrelevant features. This dynamic control mechanism enables the model to adaptively adjust its focus on different risk feature dimensions based on real-time changes in risk. During periods of high risk volatility, the model automatically increases its attention to high-uncertainty features, improving warning sensitivity. During periods of relatively stable risk, it reduces the processing of redundant features, improving computational efficiency. In this way, the binary risk gate effectively enhances the model's adaptability to complex risk scenarios and its predictive accuracy.

[0056] In one embodiment, the multimodal risk data includes policyholder behavior sequences, geographic area risk indicators, and industry claim rates.

[0057] In one embodiment, when acquiring multimodal risk data of insurance business including time, space, and industry dimensions, a hierarchical data transmission network is constructed; dedicated communication queue pairs are allocated for risk data of different sources and different importance, and a meta-information tag including data type, update time, and risk level identifier is generated for each data block; During the data transmission process, the transmission priority of the communication queue pair is dynamically adjusted according to the risk level identifier in the metadata tag. At the same time, the data integrity is monitored in real time through the metadata verification mechanism. Once a data anomaly is detected, the corresponding data block is immediately requested to be retransmitted from the original data source.

[0058] In this embodiment, in the insurance business scenario, multimodal risk data includes various types, such as policyholder behavior sequences, geographic risk indicators, and industry claims rates. Policyholder behavior sequences record the chronological order and specific actions of policyholders, such as policy applications, policy cancellations, and claims settlements. Geographic risk indicators cover spatially related risk factors such as the frequency of natural disasters and economic development levels in different regions. Industry claims rates reflect the proportion of insurance claims paid out across various industries. These data come from a wide range of sources and vary in importance. To ensure efficient data transmission, a hierarchical data transmission network is constructed, assigning dedicated communication queue pairs to risk data of varying sources and importance. These dedicated communication queue pairs act as "dedicated channels" for data transmission, ensuring that different types of data do not interfere with each other. Furthermore, a metadata tag is generated for each data block. The tag includes the data type (identifying whether it is a policyholder behavior sequence, geographic risk indicator, or industry claims rate), the update time (recording the time of the most recent data update), and the risk level identifier (classifying the risk level reflected by the data). This operation lays the foundation for subsequent data transmission management and monitoring.

[0059] During the data transmission process, the risk level identification in the meta-information tag plays a key role. The system dynamically adjusts the transmission priority of the communication queue pair based on this identification. For data with a high risk level, it assigns it a higher transmission priority so that it can be transmitted through the communication queue pair first, ensuring that important risk information can be delivered to the processing end in a timely manner, avoiding untimely risk response due to transmission delays. At the same time, the metadata verification mechanism is used to monitor data integrity in real time. The metadata verification mechanism will check whether there is data loss, error or tampering during the data transmission process. Once a data anomaly is detected, the system will immediately request the retransmission of the corresponding data block from the original data source to ensure that the acquired multimodal risk data is accurate and complete, and provide reliable data support for subsequent dynamic risk assessment, real-time risk warning and other operations based on this data.

[0060] In one embodiment, after generating a real-time warning signal based on the target risk characteristics output by the decoder and a preset risk level threshold, the process includes: Extract key characters from the multimodal risk data of each dimension and sequentially add the key characters to the nodes of the corresponding undirected graph to obtain a time dimension undirected graph, a space dimension undirected graph, and an industry dimension undirected graph; add one character to each node; Based on the characteristics of the time dimension undirected graph, the space dimension undirected graph is mutated to obtain a mutated space dimension undirected graph; based on the common properties of the time dimension undirected graph and the space dimension undirected graph, the industry dimension undirected graph is structurally adjusted to obtain an adjusted industry dimension undirected graph; generating a warning level based on the real-time warning information, and generating a first visual information carrier according to the warning level; generating a first character string based on the first visual information carrier and the time dimension undirected graph, and generating a second visual information carrier based on the first character string; generating a second character string based on the second visual information carrier and the variation space dimension undirected graph, and generating a third visual information carrier based on the second character string; An encrypted password string is generated based on the third visual information carrier and the adjusted industry dimension undirected graph, and the multimodal risk data is encrypted and stored in the insurance business database; wherein, the first visual information carrier, the second visual information carrier and the third visual information carrier are visual information carriers of different forms.

[0061] In this embodiment, in the insurance business, multimodal risk data covers multiple dimensions such as time, space, and industry. The time dimension includes time series data such as the time of insurance and the time of claim settlement; the space dimension involves spatial information such as the region where the policyholder is located and the geographical location of the insurance subject; the industry dimension includes the industry to which the policyholder belongs and the industry attributes corresponding to the insurance product. For these data of different dimensions, the key characters are first extracted. The key characters can be numbers in the timestamp, keywords in the region name, core words in the industry name, etc. Then, these key characters are added to the corresponding undirected graph nodes in turn. The nodes of the undirected graph represent the smallest information unit in the data. Only one character is added to each node to construct a time dimension undirected graph, a space dimension undirected graph, and an industry dimension undirected graph. This step converts the abstract multimodal risk data into a structured graphical representation, which is convenient for subsequent analysis and processing.

[0062] The time-dimensional undirected graph records the temporal characteristics of risk data. Based on these characteristics, a mutation operation is performed on the spatial-dimensional undirected graph. This adjustment includes mutation operations such as node splitting or merging, resulting in a mutated spatial-dimensional undirected graph. The mutation rules are customizable. Next, the common attributes of the time-dimensional and spatial-dimensional undirected graphs are analyzed. These common attributes can be shared by the structural properties of the undirected graphs. Based on these common attributes, the industry-dimensional undirected graph is structurally adjusted. The adjustment rules are customizable. Through mutation and structural adjustment of the undirected graph, the uniqueness of the undirected graph is enhanced, making it more difficult to decipher.

[0063] Furthermore, based on the target risk characteristics output by the decoder and the real-time warning signal generated by the preset risk level threshold, the warning signal is interpreted and analyzed to determine the specific warning level, which can be categorized as high risk, medium risk, and low risk. Based on the determined warning level, the relevant risk information is visualized using charts, heat maps, line graphs, and other forms to intuitively present key information such as the risk level, the time range of risk occurrence, spatial distribution, and the industries involved, thereby generating a first visual information carrier. This visual information carrier can be an information carrier such as a graph, curve, matrix, array, or data table.

[0064] Furthermore, the key content in the first visual information carrier, such as the risk level and the time point at which the major risk occurred, is combined with the node character information in the time-dimension undirected graph and arranged in a certain logical order to generate a first character string. For example, the time node characters corresponding to the high risk level are sequentially connected with the characters of the relevant time nodes in the time-dimension undirected graph. The first character string is then further visualized using a form different from the first visual information carrier, such as a text list or a tree diagram, to clearly display the information in the first character string, generating a second visual information carrier. This process achieves the transformation from graphical information to character information and then to new visual information, presenting risk information from another perspective.

[0065] Furthermore, the key data in the second visual information carrier is combined with the node character information in the undirected graph of the variation space dimension, and the characters are arranged and combined according to specific rules to generate a second character string. For example, the risk area information involved in the second visual information carrier is combined with the characters of the corresponding regional nodes in the undirected graph of the variation space dimension. Subsequently, the second character string can also be visualized using different visualization forms such as map annotation and matrix diagrams, which are different from the previous two, to obtain a third visual information carrier. This third visual information carrier further enriches the display dimension of risk information and presents risks in a more detailed way from a spatial perspective.

[0066] Finally, the key information from the third visual information carrier is integrated with the node character information in the adjusted industry dimension undirected graph and arranged according to established encoding rules to generate an encrypted password string. This encoding rule can include customized character replacement, position transformation, and other methods. The generated encrypted password string is used to encrypt the original multimodal risk data to ensure data security. Once encrypted, the encrypted data is stored in the insurance business database for subsequent query, analysis, and backtracking. This entire process, through multi-dimensional data processing, visual conversion, and encrypted storage, achieves comprehensive management and security protection of insurance business risk data.

[0067] In one embodiment, after generating a real-time warning signal based on the target risk characteristics output by the decoder and a preset risk level threshold, the process includes: Mapping the dimensional features of the multimodal risk data into a three-dimensional index table, where each index item corresponds to a eigenvalue; wherein the time dimension constructs a time dimension index, the space dimension constructs a space dimension index, and the industry dimension constructs an industry dimension index; each index item contains three attributes: eigenvalue, confidence score, and timestamp, forming a three-dimensional index table; Based on the temporal changes of the time dimension index, the spatial dimension index is regionally clustered to generate a spatiotemporal correlation matrix; By indexing the business types in the industry dimension, the spatiotemporal correlation matrix is ​​expanded into a three-dimensional risk propagation network; Encoding the real-time warning signal into a thermal value matrix; converting the thermal value matrix into a risk thermal distribution map through color gradient mapping to generate a visual information carrier; Generate an initial feature code based on the visual information carrier and the time dimension index; Performing matrix transformation on the initial feature code and the spatiotemporal correlation matrix to generate a second feature code with spatiotemporal characteristics; The business type code indexed by the industry dimension is used to verify the second feature code, generate the final risk feature key, encrypt the multimodal risk data and store it in the insurance business database.

[0068] In this embodiment, first, the dimensional features of the multimodal risk data are structured and mapped to construct a three-dimensional index table. In the time dimension, a time dimension index is established in chronological order to accurately record the timestamp of each risk data, such as the insurance time, claim time, etc.; in the spatial dimension, a spatial dimension index is constructed based on geographic coordinates or administrative divisions to mark the specific location corresponding to the data; in the industry dimension, an industry dimension index is established according to different insurance business types to distinguish between business categories such as health insurance and auto insurance. Each index item contains three attributes: eigenvalue, confidence score and timestamp. The eigenvalue quantifies the risk data, the confidence score evaluates the reliability of the data, and the timestamp clarifies the time when the data was generated. In this way, the originally scattered multimodal data is integrated into an orderly three-dimensional index table, laying the foundation for subsequent analysis.

[0069] Next, based on the temporal variations revealed by the time-dimensional index, the spatial-dimensional index is clustered. This automatically identifies fluctuation trends and cyclical characteristics in the risk data within the time dimension, such as a persistent increase in claims settlement rates in a specific region over a certain period of time. Based on this, regions with similar risk characteristics are grouped together within the spatial-dimensional index. The similarity of risk characteristics between regions is calculated to generate a spatiotemporal correlation matrix. This matrix, with time as the horizontal axis and spatial regions as the vertical axis, reflects the degree of correlation between risk data under different spatiotemporal combinations, achieving a preliminary fusion of risk information across the temporal and spatial dimensions.

[0070] Subsequently, the spatiotemporal correlation matrix was expanded by incorporating business type information from the industry dimension index. For each spatiotemporal combination in the matrix, the corresponding insurance business type was associated with it, and the risk performance of different business types under these spatiotemporal conditions was analyzed. Combining historical data with industry risk transmission patterns, the risk propagation coefficients between each business type were calculated, expanding the original two-dimensional spatiotemporal correlation matrix into a three-dimensional risk propagation network that incorporates the industry dimension. This network clearly demonstrates how risks propagate over time, across space, and across industries, providing a comprehensive perspective on risk dynamics.

[0071] The real-time warning signals are then numerically encoded and converted into a thermal value matrix. Based on a pre-defined risk level-value mapping rule, each spatiotemporal region is assigned a corresponding thermal value, with the value representing the degree of risk. Using a pre-defined color gradient mapping relationship, the thermal value matrix is ​​converted into an intuitive risk thermal distribution map, using different colors to distinguish risk levels (e.g., red for high-risk areas and green for low-risk areas), creating a visual information carrier. This allows insurance professionals to quickly and intuitively understand the spatiotemporal distribution of risks.

[0072] Based on the generated visualization information carrier and time dimension index, key risk feature information is extracted and an initial feature code is generated according to specific coding rules. The initial feature code includes key location information in the risk heat distribution map, risk level, and key time points in the time dimension, initially integrating visualization information and time dimension data.

[0073] Next, a matrix transformation operation is performed on the initial signature code and the spatiotemporal correlation matrix. Through matrix multiplication and addition, the risk correlation information in the spatiotemporal dimension is integrated into the initial signature code, further strengthening the expression of the spatiotemporal characteristics in the signature code and generating a second signature code with spatiotemporal characteristics. This signature code not only retains the core information of the initial signature code but also incorporates the risk correlation characteristics in the spatiotemporal dimension, more comprehensively reflecting the spatiotemporal attributes of risk.

[0074] Finally, the second signature code is verified using the business type encoding rules indexed by the industry dimension. A verification algorithm matches the business type code with the second signature code to ensure the accuracy and integrity of the signature code. Once verified, a final risk signature key is generated. This key is used to encrypt the original multimodal risk data, which is then stored in the insurance business database. This completes the risk data processing, analysis, and secure storage process, ensuring data security and subsequent traceability.

[0075] Reference Figure 2 In another embodiment of the present invention, a real-time risk warning system for insurance business under dynamic risk assessment is provided, comprising: The acquisition module is used to obtain multimodal risk data in insurance business, including time, space, and industry dimensions; A generation module is used to separate the core risk element area from the multimodal risk data through a mask mechanism and generate an initial risk feature map with mask labels; An input module, which feeds the initial risk profile into a dynamic risk assessment network based on an encoder-decoder architecture. Both the encoder and decoder contain multi-stage cascaded risk processing modules, each of which integrates a gated recurrent unit and a binary risk gate to dynamically control the update path of the risk profile. The early warning module is used to generate a real-time early warning signal based on the target risk characteristics output by the decoder and the preset risk level threshold.

[0076] In this embodiment, for the specific implementation of each module in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0077] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0078] Those skilled in the art will understand that Figure 3The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0079] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0080] In summary, the real-time risk warning method and system for insurance business under dynamic risk assessment provided in the embodiment of the present invention include: obtaining multimodal risk data containing time, space, and industry dimensions in the insurance business; separating the core risk element area from the multimodal risk data through a masking mechanism to generate an initial risk feature map with mask markings; inputting the initial risk feature map into a dynamic risk assessment network of an encoder-decoder architecture; wherein the encoder and decoder both include multi-stage cascaded risk processing modules, each risk processing module integrates a gated loop unit and is configured with a binary risk gate for dynamically controlling the update path of the risk feature; based on the target risk feature output by the decoder, a real-time warning signal is generated in combination with a preset risk level threshold. In the present invention, multimodal risk data containing time, space, and industry dimensions in the insurance business are obtained, and real-time risk warnings for the insurance business are performed from multiple dimensions; at the same time, through the special network structure of the dynamic risk assessment network, the processing path of the risk feature can be dynamically adjusted, thereby improving the timeliness and accuracy of the risk warning.

[0081] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0082] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0083] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A real-time risk warning method for insurance business under dynamic risk assessment, characterized by: The following steps are involved: Acquire multimodal risk data in insurance business, including time, space, and industry dimensions; The core risk element area is separated from the multimodal risk data through a masking mechanism to generate an initial risk feature map with mask marks; The initial risk feature map is input into the dynamic risk assessment network of the encoder-decoder architecture; Both the encoder and decoder contain multi-stage cascaded risk processing modules. Each risk processing module integrates a gated recurrent unit and is configured with a binary risk gate to dynamically control the update path of risk features. Based on the target risk characteristics output by the decoder and combined with the preset risk level threshold, a real-time warning signal is generated.

2. The real-time risk warning method for insurance business under dynamic risk assessment according to claim 1 is characterized in that: The encoder is used to perform time-dimensional mixed mapping on the initial risk feature map, including: After normalizing the initial risk feature map, the target update probability of the current window is calculated based on the state update probability and change value of the previous window through the multi-stage time window iteration mechanism of the gated recurrent unit; The target update probability is rounded to generate the gating parameters of the binary risk gate. The gate parameters are used to determine whether to update the hidden state features or keep the original values. After full connection processing, the temporal risk features are output; The encoder is also used for spatial dimension mixed mapping, including: Split the time series risk characteristics into regional risk sub-characteristics and industry risk sub-characteristics along the risk dimension channel; Perform convolution operations on the regional risk sub-features along the time axis in the sub-dimensions to generate regional risk propagation features; The regional transmission characteristics and industry risk sub-characteristics are integrated through element-by-element product to obtain the risk characteristic matrix that represents the spatiotemporal correlation. The decoder performs multi-scale feature reconstruction on the risk feature matrix representing spatiotemporal correlation to obtain the target risk feature.

3. The real-time risk warning method for insurance business under dynamic risk assessment according to claim 1 is characterized in that: The training process of the dynamic risk assessment network includes: Construct training samples; Extract features from training samples through a dynamic risk assessment network to generate risk feature prediction distribution; The variance matrix of the predicted distribution is calculated, and the target loss function is constructed by combining the difference between the predicted distribution and the risk label and the variance matrix to perform end-to-end training on the dynamic risk assessment network.

4. The real-time risk warning method for insurance business under dynamic risk assessment according to claim 3 is characterized in that: The constructing of the target loss function includes: In the first stage, the accuracy of the model on risk variance is constrained based on the differences in spatiotemporal coordinates and variance matrix between the predicted distribution and the risk label; In the second stage, weights are assigned to features of different dimensions based on risk variance to strengthen feature learning in high uncertainty areas; The dynamic risk assessment network is trained sequentially based on two-stage loss functions.

5. The real-time risk warning method for insurance business under dynamic risk assessment according to claim 4 is characterized in that: The variance matrix includes time dimension variance, space dimension variance and time-space coupling variance. The accuracy of the constraint model on risk variance includes: Calculate the characteristic differences between the predicted distribution and the risk label in time, space, and spatiotemporal coupling dimensions; The first risk loss term is obtained by weighting the feature differences with a function whose exponent is the variance of each dimension; Determine the second risk loss item based on the risk level distribution difference and variance matrix; The objective loss function of the first stage is obtained by weighted summing of the first risk loss term and the second risk loss term.

6. The real-time risk warning method for insurance business under dynamic risk assessment according to claim 2 is characterized in that: The dynamic control mechanism of the binary risk gate includes: For the current time window, the target update probability matrix is ​​predicted through the logistic regression model based on the state update probability sequence and change gradient of the previous window; The target update probability matrix is ​​binarized to generate a gating mask, and the attention head of the gated recurrent unit is selectively activated according to the gating mask to achieve adaptive weighted aggregation of temporal risk features.

7. The real-time risk warning method for insurance business under dynamic risk assessment according to claim 1 is characterized in that: The multimodal risk data includes policyholder behavior sequences, geographical area risk indicators, and industry claim rates.

8. A real-time risk warning system for insurance business under dynamic risk assessment, characterized by: include: The acquisition module is used to obtain multimodal risk data in insurance business, including time, space, and industry dimensions; A generation module is used to separate the core risk element area from the multimodal risk data through a mask mechanism and generate an initial risk feature map with mask labels; An input module, used to input the initial risk feature map into the dynamic risk assessment network of the encoder-decoder architecture; Both the encoder and decoder contain multi-stage cascaded risk processing modules. Each risk processing module integrates a gated recurrent unit and is configured with a binary risk gate to dynamically control the update path of risk features. The early warning module is used to generate a real-time early warning signal based on the target risk characteristics output by the decoder and the preset risk level threshold.

Images