Life insurance reinsurance scheme generation method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of financial science and technology, can be applied to the field of reinsurance, and discloses a life insurance reinsurance scheme generation method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring data related to a target life insurance reinsurance service from at least two data sources as multi-source life insurance reinsurance data; performing data cleaning processing on the multi-source life insurance reinsurance data, and extracting and fusing multi-dimensional features from the multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain life insurance reinsurance risk feature vectors; according to the life insurance reinsurance risk feature vector, constructing a life insurance reinsurance dynamic risk assessment model, and outputting a life insurance reinsurance risk assessment result through the life insurance reinsurance dynamic risk assessment model; and on the basis of a life insurance reinsurance risk assessment result, a life insurance reinsurance allocation scheme is generated through collaborative optimization of a multi-objective optimization algorithm and deep reinforcement learning in combination with a real-time triggering mechanism. According to the method, the rationality, accuracy and dynamic adaptability of a life insurance reinsurance scheme are improved.

Description

Technical Field

[0001] This application relates to the field of financial technology and can be applied to the reinsurance field, particularly to a method, apparatus, electronic device and storage medium for generating life insurance reinsurance schemes. Background Technology

[0002] In life reinsurance, traditional risk assessment and reinsurance scheme development processes rely heavily on structured historical data, such as policy information and past claims records, combined with static actuarial models (e.g., statistical models based on fixed mortality and morbidity assumptions) for risk calculation. In the reinsurance allocation stage, methods like linear programming, simple priority ranking, or fixed-ratio allocation are typically used to generate schemes. This means that risks are distributed among different entities according to pre-defined rules based on limited conditions such as the reinsurer's capital size and cooperation history. However, this model has gradually revealed significant limitations in practical applications: Firstly, risk assessment relies excessively on a single structured data source, making it difficult to integrate unstructured information such as medical images and health report texts. Furthermore, static models cannot respond in real-time to dynamic environmental changes such as economic fluctuations and public health emergencies, resulting in insufficient granularity of risk identification. Secondly, reinsurance allocation algorithms are prone to getting trapped in local optima when dealing with multiple objectives such as cost control, risk diversification, and compliance constraints. This is especially true when faced with a large number of reinsurers to choose from, leading to low computational efficiency and difficulty in generating a globally optimal solution. At the same time, data silos exist between insurance companies, reinsurers, and third-party institutions, information sharing is not smooth, and there is a lack of real-time feedback mechanisms, which makes it impossible to quickly adjust the plan when the reinsurer's credit rating changes or when there are sudden market fluctuations. Summary of the Invention

[0003] The main technical problem addressed by the implementation method of this application is the low accuracy and efficiency of risk identification when generating life insurance reinsurance plans using traditional technologies.

[0004] To address the aforementioned technical problems, the first technical solution adopted in this application is: providing a method for generating a life insurance reinsurance scheme, comprising: acquiring structured data, unstructured data, and real-time data related to the target life insurance reinsurance business from at least two preset data sources, as multi-source life insurance reinsurance data; performing data cleaning processing on the multi-source life insurance reinsurance data, extracting and fusing multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector; constructing a life insurance reinsurance dynamic risk assessment model including a temporal convolutional network and a dynamic attention mechanism based on the life insurance reinsurance risk feature vector, outputting a life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model; and generating a life insurance reinsurance allocation scheme based on the life insurance reinsurance risk assessment result, using the risk-bearing ratio of each reinsurance company as the decision variable, through multi-objective optimization algorithm and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism.

[0005] Optionally, the step of extracting and fusing multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector includes: extracting text feature vectors from text data in the multi-source life insurance reinsurance data using a pre-trained bidirectional encoder-representation converter model; extracting image feature vectors from image data in the multi-source life insurance reinsurance data using a residual network; constructing a reinsurance company association graph using a graph neural network, aggregating neighborhood information based on the reinsurance company association graph to obtain reinsurance company rating data; inputting dynamic data from the multi-source life insurance reinsurance data into a temporal convolutional network to generate time-sensitive risk trend feature data; and aligning the feature dimensions of the text feature vector, image feature vector, reinsurance company rating data, and risk trend feature data through a cross-attention mechanism, and fusing them to generate a unified life insurance reinsurance risk feature vector.

[0006] Optionally, the step of constructing a dynamic risk assessment model for life reinsurance that includes a temporal convolutional network and a dynamic attention mechanism based on the life reinsurance risk feature vector includes: constructing a temporal convolutional network using the life reinsurance risk feature vector as input, and using multi-layer dilated causal convolution to capture long-term temporal dependencies; setting target parameters for the dilated convolutional layers of the temporal convolutional network, wherein the kernel size is set to a preset first number, and the dilation factor is set to grow exponentially; connecting a dynamic attention mechanism after the temporal convolutional network, and calculating the first attention weights for each time step of the hidden state output by the temporal convolutional network; performing a linear transformation on the hidden state of the current time step and the historical state matrix through a weight matrix, wherein the historical state matrix is ​​generated by the temporal convolutional network after performing multi-layer dilated causal convolution on the input temporal features; calculating the similarity score between the transformed hidden state of the current time step and the historical state matrix, and normalizing the similarity score to obtain the second attention weights; and generating a dynamic risk feature vector by weighted summation of the historical state matrix based on the second attention weights, thereby obtaining the constructed dynamic risk assessment model for life reinsurance.

[0007] Optionally, before the step of outputting the life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model, the method further includes: designing a loss function for multi-task joint training, wherein the main task of the loss function is to calculate the first loss of the expected payout probability of an individual in the next preset second number of years through binary classification cross-entropy, and the auxiliary task of the loss function is to calculate the second loss of the life insurance reinsurance risk due to preset volatility data through mean squared error loss; using a course learning training strategy, the life insurance reinsurance dynamic risk assessment model is trained through the multi-source life insurance reinsurance data in descending order of confidence level, to obtain the trained life insurance reinsurance dynamic risk assessment model.

[0008] Optionally, the step of generating a life insurance reinsurance allocation scheme based on the life insurance reinsurance risk assessment results, using the risk-bearing ratio of each reinsurance company as the decision variable, and through multi-objective optimization algorithms and deep reinforcement learning collaborative optimization combined with a real-time triggering mechanism, includes: setting the risk-bearing ratio of each reinsurance company as the decision variable; constructing an objective function that includes cost minimization and risk diversification maximization; the cost minimization is used to comprehensively calculate the product of the reinsurance premium rate of each reinsurance company and the corresponding risk ratio, the product of the risk penalty term and the corresponding risk value, and then summing them to obtain the total cost, so as to solve for the minimization of the total cost; the risk diversification maximization is used to maximize the degree of risk diversification by allocating the risk ratio of each reinsurance company; setting constraints, The constraints include that the risk proportion borne by a single reinsurance company does not exceed a preset first risk proportion threshold, and the total risk proportion borne by all reinsurance companies is 1; an improved multi-objective evolutionary algorithm and deep reinforcement learning are used for collaborative optimization; the improved multi-objective evolutionary algorithm randomly generates initial solutions that satisfy the constraints, performs adaptive crossover and mutation operations on parent solutions, performs non-dominated sorting on the solution set, and retains Pareto front solutions; the deep reinforcement learning adjusts the crossover and mutation parameters or population selection strength parameters of the multi-objective evolutionary algorithm based on the characteristics of the current solution set; a real-time triggering mechanism is established, and when the dynamic changes in data in a preset data dimension exceed a preset threshold, a local re-optimization process is triggered, retaining the original solutions with a preset proportion as the initial population to accelerate convergence and generate the life insurance reinsurance allocation scheme.

[0009] Optionally, the method for generating a life insurance reinsurance scheme further includes: constructing a private blockchain and setting node roles, wherein the insurance company acts as the initiating node, the reinsurance company acts as the verification node, and the regulatory agency acts as the audit node; creating smart contract logic to automatically verify whether the life insurance reinsurance allocation scheme meets preset verification rules; and if the verification is successful, uploading the life insurance reinsurance risk assessment result and the hash value of the life insurance reinsurance allocation scheme to the private blockchain for storage.

[0010] Optionally, the life insurance reinsurance scheme generation method further includes: obtaining feedback data of the life insurance reinsurance allocation scheme, the feedback data including the reinsurance company's actual loss ratio data, interest rate adjustment data, and risk-related data; setting the feedback data as new training data, using an incremental learning mechanism to iteratively optimize the life insurance reinsurance dynamic risk assessment model, and retaining the old weights of the selected parameters in the life insurance reinsurance dynamic risk assessment model through an elastic weight consolidation algorithm; and updating the parameters of the life insurance reinsurance dynamic risk assessment model based on the new training data to continuously optimize the life insurance reinsurance dynamic risk assessment model.

[0011] To address the aforementioned technical problems, the second technical solution adopted in this application is: providing a life insurance reinsurance scheme generation device, comprising: a business data acquisition module, used to acquire structured data, unstructured data, and real-time data related to the target life insurance reinsurance business from at least two preset data sources, as multi-source life insurance reinsurance data; a data cleaning and processing module, used to perform data cleaning processing on the multi-source life insurance reinsurance data, extracting and fusing multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector; a risk assessment result module, used to construct a life insurance reinsurance dynamic risk assessment model including a temporal convolutional network and a dynamic attention mechanism based on the life insurance reinsurance risk feature vector, and output a life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model; and a life insurance reinsurance scheme module, used to generate a life insurance reinsurance allocation scheme based on the life insurance reinsurance risk assessment result, using the risk-bearing ratio of each reinsurance company as the decision variable, through multi-objective optimization algorithm and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism.

[0012] To solve the above-mentioned technical problems, the third technical solution adopted in the embodiments of this application is: to provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the life insurance reinsurance scheme generation method as described above.

[0013] To solve the above-mentioned technical problems, the fourth technical solution adopted in the embodiments of this application is: to provide a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by an electronic device, the electronic device executes the life insurance reinsurance scheme generation method as described above.

[0014] Unlike related technologies, the life insurance reinsurance scheme generation method provided in this application overcomes the limitations of a single data source by integrating multi-source heterogeneous data; it solves the problem of data silos by generating a unified risk feature vector through a multi-dimensional feature extraction mechanism; it constructs an evaluation model that integrates temporal convolutional networks and dynamic attention mechanisms, which can accurately capture long-term risk trends and short-term fluctuations; and it uses the reinsurance company's risk-bearing ratio as a decision variable, and through multi-objective optimization and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism, the generated reinsurance scheme takes into account both cost control and risk diversification, and can dynamically adapt to changes, effectively improving the scientificity, accuracy, and adaptability of life insurance reinsurance schemes. Attached Figure Description

[0015] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0016] Figure 1 This is a schematic diagram of the operating environment of the life insurance reinsurance scheme generation method provided in the embodiments of this application.

[0017] Figure 2 This is a schematic diagram of the execution flow of the life insurance reinsurance scheme generation method provided in the embodiments of this application.

[0018] Figure 3 This is a schematic diagram of the execution flow for generating life insurance reinsurance risk feature vectors in the life insurance reinsurance scheme generation method provided in this application embodiment.

[0019] Figure 4 This is a schematic diagram of the execution flow of constructing a dynamic risk assessment model for life insurance reinsurance in the life insurance reinsurance scheme generation method provided in the embodiments of this application.

[0020] Figure 5 This is a schematic diagram of the system structure of the life insurance reinsurance scheme generation device provided in the embodiments of this application.

[0021] Figure 6 This is a schematic diagram of the hardware structure of the electronic device for executing the life insurance reinsurance scheme generation method provided in the embodiments of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Software tools, components, or servers not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0023] It should be noted that, unless otherwise specified, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a 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 schematic diagram or the order in the flowchart.

[0024] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0025] To facilitate understanding of this embodiment, a method for generating a life insurance reinsurance plan disclosed in this application will first be described in detail. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a schematic diagram of the operating environment for the life insurance reinsurance scheme generation method provided in the embodiments of this application, such as... Figure 1 As shown, the execution subject of the life insurance reinsurance plan generation method provided in this application embodiment is generally an electronic device with a certain computing power, such as a computer device. In some possible implementations, this life insurance reinsurance plan generation method can be implemented by a processor calling computer-readable instructions stored in memory. Figure 1 The computer equipment mentioned can be a server. A server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. This can be understood as... Figure 1 The number of computer devices shown is merely illustrative and can be expanded in any number according to actual needs.

[0026] Please continue reading. Figure 2 , Figure 2 This is a schematic diagram of the execution flow of the life insurance reinsurance scheme generation method provided in the embodiments of this application, such as... Figure 2 As shown, it includes the following steps:

[0027] S1. Obtain structured data, unstructured data, and real-time data related to the target life insurance reinsurance business from at least two preset data sources, as multi-source life insurance reinsurance data.

[0028] Life reinsurance, also known as life insurance reinsurance business, refers to a business model whereby an insurance company, in order to diversify the risks of its life insurance business, transfers a portion of the liability of its underwritten life insurance policies to other reinsurance companies (i.e., reinsurers) through reinsurance contracts. The reinsurers then assume the corresponding risk liability under agreed-upon conditions. This business helps insurance companies reduce the impact of single risks, enhance their risk resilience, and ensure the stable operation of their business. When acquiring multi-source life insurance reinsurance data, the data sources involved are quite extensive, including insurance companies' internal business systems, reinsurer risk assessment report databases, publicly available data platforms of industry regulatory authorities, and macroeconomic and population health databases provided by third-party data service providers. Structured data refers to data with a fixed format that is easy to store and retrieve, such as information like the policyholder's age, gender, sum insured, premium amount, and payment period in a life insurance policy, as well as the reinsurance company's registered capital, credit rating, and historical claims records. Unstructured data, on the other hand, is data without a fixed format, such as the policyholder's medical records, health examination images (e.g., CT scans, ultrasound images), meeting minutes related to reinsurance business, and industry analysis reports. Real-time data refers to data that changes dynamically over time and needs to be obtained promptly, such as real-time updated capital market interest rates, real-time statistics on the incidence of critical illnesses, and immediate adjustments to regulatory policies. The aforementioned multi-source life insurance reinsurance data forms the basis for subsequent life insurance reinsurance scheme generation. Structured data provides clear quantitative evidence for risk assessment, unstructured data supplements potential risk information, and real-time data reflects the dynamic changes in the current market and business environment. The combination of these three can comprehensively depict the risk characteristics of life insurance reinsurance business, providing strong data support for accurate risk assessment, model building, and reinsurance scheme optimization.

[0029] S2. Perform data cleaning on the multi-source life insurance reinsurance data. Extract and fuse multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain the life insurance reinsurance risk feature vector.

[0030] As an alternative implementation method, please continue reading. Figure 3 , Figure 3 This is a schematic diagram of the execution flow of generating life insurance reinsurance risk feature vectors in the life insurance reinsurance scheme generation method provided in this application embodiment, as shown below. Figure 3 As shown, the process can specifically include the following steps S21 to S25.

[0031] S21. Extract text feature vectors from text data in multi-source life insurance reinsurance data using a pre-trained bidirectional encoder-representation converter model.

[0032] Among them, bidirectional encoder representation transformer models (such as the BERT model) are based on a bidirectional Transformer structure, capable of simultaneously capturing the semantic relationships within the context of text, making them suitable for transforming unstructured text into structured feature vectors. In life insurance reinsurance business, textual data may include the insured's medical records (e.g., "no history of major illnesses in the past 5 years, blood pressure slightly above the normal range"), liability clauses in reinsurance contracts (e.g., "the payout ratio for malignant tumors is no less than 80%)", risk warning documents issued by regulatory agencies, etc. After extracting feature vectors through this model, implicit risk information in the text (e.g., potential health risks in medical records, liability boundaries in clauses) can be transformed into quantitative signals that the model can recognize, providing semantic support for subsequent risk assessment.

[0033] S22. Extract image feature vectors from image data in multi-source life insurance reinsurance data using residual networks.

[0034] Among them, ResNet addresses the vanishing gradient problem in deep neural network training by introducing skip connections, effectively extracting deep visual features of images (such as texture, structure, and key regions). In life insurance reinsurance scenarios, image data may include the insured's medical examination images (e.g., chest CT scans, echocardiograms), scanned copies of the reinsurance company's financial statements (including charts), and geographical distribution maps of critical illness incidence rates. For example, on the insured's chest CT scans, ResNet can automatically identify the presence of abnormal features such as nodules and calcifications, and transform this visual information into feature vectors, complementing the health descriptions in the text medical records and improving the comprehensiveness of individual health risk assessment.

[0035] S23. Use a graph neural network to construct a reinsurance company association graph, and aggregate neighborhood information based on the reinsurance company association graph to obtain reinsurance company scoring data.

[0036] Graph Neural Networks (GNNs) excel at processing graph-structured data with interconnected relationships. They quantify the overall value of nodes by aggregating neighborhood information (e.g., neighboring node attributes and connection strength) to update node features. In life reinsurance, the nodes in the reinsurance company graph represent the individual reinsurance companies themselves. Edge weights can indicate the depth of cooperation between companies (e.g., the number of jointly underwritten policies and the length of historical cooperation) and the degree of risk correlation (e.g., the proportion of jointly undertaken high-risk policies). For example, if reinsurance company A has long-term, deep cooperation with three reinsurance companies with AA+ credit ratings, the GNN will improve company A's rating by aggregating the positive attributes of these three companies. Conversely, if a reinsurance company's neighboring nodes are mostly high-default-risk entities, its rating will decrease accordingly. This rating data directly reflects the reinsurance company's risk-bearing capacity and the reliability of its cooperation.

[0037] S24. Input the dynamic data from the multi-source life insurance reinsurance data into the temporal convolutional network to generate time-sensitive risk trend feature data.

[0038] Temporal Convolutional Networks (TCNs) employ dilated causal convolutional structures, effectively capturing long-term temporal dependencies (e.g., quarterly or annual trend changes) when processing time-series data, while avoiding the gradient explosion problem of recurrent neural networks. In life insurance reinsurance scenarios, dynamic data may include monthly loss ratios (e.g., "critical illness insurance loss ratio increased by 12% in Q1 2023 compared to Q4"), quarterly economic volatility indices (e.g., GDP growth rate, inflation rate), and annual incidence rates of critical illnesses (e.g., "the annual growth rate of thyroid cancer incidence rate has been 8% in the past 3 years"). After processing by a temporal convolutional network, the temporal trend characteristics of these data can be extracted (e.g., seasonal fluctuations in loss ratios, lagged correlations between economic indicators and incidence rates), providing the model with a dynamic perspective of "risk evolving over time" and improving the ability to predict future risks.

[0039] S25. Align the feature dimensions of text feature vectors, image feature vectors, reinsurance company scoring data, and risk trend feature data through a cross-attention mechanism, and fuse them to generate a unified life insurance reinsurance risk feature vector.

[0040] Specifically, a cross-attention mechanism aligns the feature dimensions of text feature vectors, image feature vectors, reinsurance company rating data, and risk trend feature data, fusing them to generate a unified life insurance reinsurance risk feature vector. The cross-attention mechanism calculates attention weights between different features, focusing on the correlation between features (e.g., the correlation between "hypertension" in a medical record and "cardiac enlargement" in a CT scan), and mapping multi-source features with inconsistent dimensions to the same vector space. In life insurance reinsurance, features such as the policyholder's text medical record ("5-year history of hypertension"), physical examination image features ("left ventricular hypertrophy"), the rating data of the cooperating reinsurance company ("credit rating AA, strong risk-bearing capacity"), and time trend features ("hypertension-related claims rate increased by 5% annually in the past 3 years"), after being aligned and fused by the cross-attention mechanism, generate a unified risk feature vector that comprehensively reflects multi-dimensional information such as individual health risk, reinsurer capability, and time trends. This provides a comprehensive and structured input for subsequent dynamic risk assessment models, ensuring that the model can capture the complex risk characteristics of life insurance reinsurance from a holistic perspective.

[0041] Through steps S21 to S25 above, in-depth mining and efficient fusion of multi-source life insurance reinsurance data can be achieved: by using pre-trained models and residual networks to transform unstructured text and image data into quantifiable feature vectors, breaking down the barriers to the application of unstructured data; by aggregating reinsurance company association information through graph neural networks to generate scoring data, and by combining temporal convolutional networks to capture the time trend of dynamic data, the ability to capture association and dynamic features is strengthened; finally, by using cross-attention mechanisms to align and fuse multi-dimensional features, a unified risk feature vector is generated, which not only retains the unique value of each dimension but also achieves complementary enhancement, solving the problem of data "islanding" and providing comprehensive and accurate input for subsequent dynamic risk assessment models, laying the foundation for the accuracy of risk assessment.

[0042] S3. Based on the risk feature vector of life insurance reinsurance, construct a dynamic risk assessment model for life insurance reinsurance that includes a temporal convolutional network and a dynamic attention mechanism, and output the life insurance reinsurance risk assessment result through the trained dynamic risk assessment model for life insurance reinsurance.

[0043] As an alternative implementation method, please continue reading. Figure 4 , Figure 4 This is a schematic diagram of the execution flow of constructing a dynamic risk assessment model for life reinsurance in the life reinsurance scheme generation method provided in this application embodiment, as shown below. Figure 4 As shown, the process can specifically include the following steps S31 to S36.

[0044] S31. Using the life insurance reinsurance risk feature vector as input, construct a temporal convolutional network and use multi-layer dilated causal convolution to capture long-term temporal dependencies.

[0045] Among them, Temporal Convolutional Networks (TCNs) ensure unidirectional processing of time-series data (predicting the future using only past data) through causal convolution, while dilated convolutions expand the receptive field by inserting zero values ​​between convolution kernel elements, and multiple layers can cover longer time series. In life insurance reinsurance scenarios, risk feature vectors contain time-series information such as the health data of insured individuals over multiple years and the history of cooperation with reinsurance companies. For example, the payout rate of a certain type of critical illness insurance shows periodic fluctuations between 2018 and 2023. Multi-layer dilated causal convolutions can capture this long-term dependency across years and avoid risk misjudgment caused by ignoring historical correlations.

[0046] S32. Set target parameters for the dilated convolutional layers of the temporal convolutional network, where the kernel size is set to a preset first number and the dilation factor is set to grow exponentially.

[0047] Among them, the size of the convolution kernel determines the width of the local time window for single-step processing. For example, a convolution kernel of size 3 can capture local features of three consecutive time steps (such as risk fluctuations within a quarter). The exponential growth of the expansion factor causes the receptive field of the model to expand exponentially with the number of layers. For example, when the expansion factors of a 4-layer network are 1, 2, 4, and 8, the total receptive field can reach 1+2+4+8=15 time steps, which is sufficient to cover 15 years of historical risk data, ensuring comprehensive capture of long-term time-series features and adapting to the risk assessment needs of long-term life insurance business (such as 10-30 years).

[0048] S33. Connect a dynamic attention mechanism after the temporal convolutional network to calculate the first attention weight of the hidden state output by the temporal convolutional network at each time step.

[0049] Among them, the dynamic attention mechanism automatically focuses on key historical nodes for current risk assessment by quantifying the importance of hidden states at different time steps.

[0050] S34. The hidden state and historical state matrix at the current time step are linearly transformed by the weight matrix. The historical state matrix is ​​generated by the temporal convolutional network after performing multi-layer dilated causal convolution on the input temporal features.

[0051] The essence of linear transformation is to map the current state and historical states to the same feature space, eliminating differences in dimensionality or distribution and laying the foundation for subsequent similarity calculations. For example, the hidden state of the current time step (2023) focuses on the risk of "aggravated aging," while the historical state matrix contains features such as "incidence rate of serious diseases" from 2018 to 2022. After transformation by the weight matrix, the correlation between the two types of states can be compared under a unified dimension.

[0052] S35. Calculate the similarity score between the hidden state at the current time step and the historical state matrix after transformation, normalize the similarity score, and obtain the second attention weight.

[0053] The similarity score (e.g., cosine similarity) measures the strength of the association between the current and historical states. Normalization (e.g., through the Softmax function) converts the score into weights that sum to 1, ensuring that the contribution of historical states can be quantified and accumulated. For example, the similarity score between the "low-interest-rate environment" risk in 2023 and 2019 is 0.8, and between 2023 and 2021 is 0.3. After normalization, the second attention weights are 0.73 and 0.27 respectively, clearly indicating the proportion of influence from historical states.

[0054] S36. Based on the weighted summation of the historical state matrix according to the second attention weight, a dynamic risk feature vector is generated, resulting in the completed dynamic risk assessment model for life insurance reinsurance.

[0055] The weighted summation process integrates features from key historical states with the current state, enabling the dynamic risk feature vector to incorporate both real-time risk information and highly correlated historical experience. For example, by combining experience in dealing with low-interest-rate risks in 2019 (weight 0.73) with real-time data from 2023, the generated feature vector can more accurately reflect the impact of current interest rate fluctuations on reinsurance risk. The final model can output a comprehensive risk assessment result that includes both long-term trends and short-term fluctuations (such as expected loss ratio and risk concentration over the next 5 years).

[0056] Through steps S31 to S36, the temporal convolutional network, through multi-layered dilated causal convolution and parameter optimization, effectively captures the long-term temporal dependencies in the risk feature vector, adapting to the long-cycle characteristics of life insurance business. The dynamic attention mechanism, through the weight calculation and transformation fusion of hidden states, accurately focuses on the correlation between key historical nodes and the current state, allowing the model to both integrate highly correlated historical risk experience and respond to the latest risk changes in real time. The resulting dynamic risk feature vector achieves an organic fusion of long-term trends and short-term fluctuations, making the risk assessment results output by the constructed model more consistent with actual business scenarios, and providing timely and forward-looking risk basis for subsequent reinsurance scheme optimization.

[0057] As another optional implementation, before the step of outputting the life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model in step S3 above, the following steps S37 to S38 may also be included.

[0058] S37. Design a loss function for multi-task joint training, wherein the main task of the loss function is to calculate the first loss of the expected payout probability of an individual in the next preset second number of years through binary classification cross-entropy, and the auxiliary task of the loss function is to calculate the second loss of the life insurance reinsurance risk of preset fluctuation data through mean squared error loss.

[0059] Multi-task joint training achieves collaborative learning of risk information from different dimensions by simultaneously optimizing the losses of primary and secondary tasks. In the life insurance reinsurance scenario, the primary task focuses on individual core risks (e.g., whether a policyholder will make a claim in the next 5 years), and binary cross-entropy can accurately quantify the deviation between the predicted probability and the actual result; the secondary task focuses on the impact of external fluctuations (e.g., the impact of market interest rate fluctuations on overall reinsurance risk), and mean squared error loss can effectively measure the difference between the predicted fluctuation impact and the actual value. The combination of the two allows the loss function to both anchor the accuracy of core risk assessment and incorporate external environmental interference factors, enabling the model to simultaneously grasp the individual risk patterns and the systemic risk transmission logic.

[0060] S38. Using the training strategy learned in the course, the dynamic risk assessment model for life insurance reinsurance is trained by using multi-source life insurance reinsurance data in descending order of confidence level, and the trained dynamic risk assessment model for life insurance reinsurance is obtained.

[0061] The learning process mimics human learning patterns, first training the model with high-confidence data (such as verified historical claims records and incidence rate data released by authoritative institutions) to master basic risk patterns, and then gradually introducing low-confidence data (such as estimated market fluctuation data and preliminary statistics of new business). In life insurance reinsurance scenarios, this strategy avoids interference from low-quality data in the early stages, allowing the model to solidify its understanding of mature risk patterns before expanding its generalization capabilities with low-confidence data. The final trained model can maintain stable predictions of core risks while adapting to complex and ever-changing business data environments, improving the reliability and robustness of assessment results.

[0062] S4. Based on the life insurance reinsurance risk assessment results, and using the risk-bearing ratio of each reinsurance company as the decision variable, a life insurance reinsurance allocation plan is generated through multi-objective optimization algorithm and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism.

[0063] As an optional implementation, step S4 may specifically include steps S41 to S4.

[0064] S41. Set the risk proportion borne by each reinsurance company as a decision variable. The decision variable is the core adjustment object in the optimization problem. In the life insurance reinsurance scenario, the risk proportion borne by each reinsurance company directly determines the risk and cost allocation structure. For example, assuming there are 3 reinsurance companies, the decision variable can be represented as (x1, x2, x3), where x1 represents the risk proportion borne by the first company, and x1 + x2 + x3 = 1. Different allocation strategies can be achieved by adjusting this variable.

[0065] S42. Construct an objective function that includes cost minimization and risk diversification maximization. Cost minimization involves comprehensively calculating the product of each reinsurance company's reinsurance rate and its corresponding risk proportion, and the product of the risk penalty term and its corresponding risk value, then summing these to obtain the total cost, thus minimizing the total cost. Risk diversification maximization involves allocating the risk proportions of each reinsurance company to achieve the highest possible overall risk dispersion, thus maximizing the degree of risk diversification.

[0066] The objective function serves as a guide for optimization. Cost minimization requires comprehensive consideration of explicit costs (the product of reinsurance rate and risk ratio; for example, a reinsurance company with a rate of 0.05 and a cost of 0.015 when bearing 30% risk) and implicit costs (the product of risk penalty and risk value; such as additional penalties for high-risk business). The total cost is obtained by summing these costs and minimizing them. Risk diversification, on the other hand, maximizes the overall risk distribution by rationally allocating risk ratios (for example, avoiding a single company bearing more than 50% of the risk). Essentially, it quantifies diversification through entropy and pursues maximization to achieve a balanced distribution of risk among multiple entities.

[0067] S43. Set constraints, including that the risk ratio borne by a single reinsurance company does not exceed the preset first risk ratio threshold, and the total risk ratio borne by all reinsurance companies is 1.

[0068] Among them, the constraint is to optimize the scope of feasibility. The former avoids the risk of single point of failure caused by excessive concentration of risk in a certain entity (for example, a reinsurance company is unable to perform its obligations due to insufficient capital), while the latter ensures that the risk is fully covered, which is in line with the basic logic of reinsurance business.

[0069] S44. Utilize an improved multi-objective evolutionary algorithm in conjunction with deep reinforcement learning for optimization. The multi-objective evolutionary algorithm excels at searching for Pareto optimal solutions in complex solution spaces, while deep reinforcement learning can dynamically adjust its algorithm strategy based on environmental feedback. The synergy between the two can balance global search capability with local optimization efficiency. In life reinsurance, this synergistic approach avoids the algorithm getting trapped in local optima (e.g., pursuing only low cost while ignoring risk concentration) and accelerates convergence to a high-quality solution that balances cost and risk.

[0070] S45. The improved multi-objective evolutionary algorithm randomly generates initial solutions that satisfy the constraints, performs adaptive crossover and mutation operations on the parent solutions, performs non-dominated sorting on the solution set, and retains Pareto front solutions.

[0071] Among these, initial solution generation ensures that the search covers a sufficiently broad solution space; adaptive crossover mutation (e.g., dynamically adjusting the mutation probability based on the diversity of solutions) can introduce new solutions while retaining high-quality genes, thus avoiding population degradation; non-dominated sorting selects Pareto front solutions that are not completely surpassed by other solutions by comparing the objective function values ​​of the solutions (e.g., if solution A has a lower cost than B and better risk diversification than B, then B is dominated), ensuring that the retained solutions are all high-quality choices after trade-offs.

[0072] S46. Deep reinforcement learning adjusts the crossover and mutation parameters or population selection strength parameters of a multi-objective evolutionary algorithm based on the features of the current solution set.

[0073] For example, when the solution set is not diverse enough, reinforcement learning increases the mutation probability to introduce new solutions; when it is close to the optimal solution, it reduces the selection strength to retain more potential high-quality solutions. This dynamic adjustment enables the algorithm to adapt to complex changes in the solution space and improve optimization efficiency and accuracy.

[0074] S47. Establish a real-time triggering mechanism. When the dynamic changes in the preset data dimension exceed the preset threshold, trigger the local re-optimization process, retain the original solution of the preset proportion as the initial population to accelerate convergence, and generate a life insurance reinsurance allocation scheme.

[0075] Among them, the real-time triggering mechanism ensures that the solution can respond quickly to external changes (such as reducing the risk-bearing ratio of a reinsurance company after its credit rating is downgraded), while retaining part of the original solution can avoid searching from scratch and use the information of the historical best solution to accelerate the finding of a new equilibrium solution, taking into account both timeliness and optimization effect.

[0076] Through steps S41 to S47 above, a life insurance reinsurance allocation plan that balances cost and risk can be efficiently generated. Using the risk-bearing ratio as the decision variable, combined with a dual objective function and constraints, a balance between cost control and risk diversification is achieved. Multi-objective evolutionary algorithms and deep reinforcement learning work together to improve efficiency, while a real-time triggering mechanism ensures the plan dynamically adapts to changes, ultimately providing the business with a scientific and flexible risk allocation strategy.

[0077] As an optional implementation method, the life insurance reinsurance scheme generation method provided in this application embodiment can also ensure the transparency and security of multi-party data sharing in life insurance reinsurance business, specifically including the following steps S51 to S53.

[0078] S51. Construct a private blockchain and set up node roles, with the insurance company as the initiating node, the reinsurance company as the verification node, and the regulatory agency as the audit node.

[0079] In this architecture, the private blockchain provides a closed and trusted distributed ledger environment for participants in life insurance reinsurance business. The differentiated roles of nodes ensure clear responsibilities: insurance companies, as initiating nodes, are responsible for uploading reinsurance business data and proposing solutions; reinsurance companies, as verification nodes, participate in consensus verification of data authenticity and solution rationality; and regulatory agencies, as auditing nodes, have full data viewing permissions to fulfill their regulatory duties. This architecture avoids the privacy risks of public blockchains and ensures that data is not tampered with by a single entity through distributed storage, laying a foundation of trust for multi-party collaboration.

[0080] S52. Create smart contract logic to automatically verify whether the life insurance reinsurance allocation plan meets the preset verification rules.

[0081] Smart contracts are automatically executable code embedded in the blockchain. In life insurance reinsurance scenarios, preset verification rules can include compliance requirements such as "a single reinsurance company's risk-bearing ratio does not exceed 40%" and "the total cost of the plan does not exceed the budget threshold." When an insurance company proposes an allocation plan, the smart contract can automatically verify whether the plan complies with the rules without human intervention. For example, if Company A bears 50% of the risk in a plan, the contract will directly determine that the verification has failed and refuse to execute it, which improves verification efficiency and avoids subjective bias and violation risks from human operation.

[0082] S53. If the verification is successful, upload the hash values ​​of the life insurance reinsurance risk assessment results and the life insurance reinsurance allocation plan to the private blockchain for storage.

[0083] Hash values ​​serve as unique digital fingerprints for data, exhibiting the characteristic that "a small change in the original data can lead to a significant change in the hash value." By storing the hash values ​​of risk assessment results (such as expected payout ratios and risk levels) and allocation schemes on the blockchain, rather than the original data, on-chain storage resources are saved while data integrity verification is achieved. At any time later, participants can quickly determine whether the data has been tampered with by recalculating the hash value and comparing it with on-chain records. For example, if a reinsurance company has doubts about the original data for a certain scheme, it can confirm the data's authenticity through hash comparison, ensuring the traceability and immutability of key information in life insurance reinsurance business.

[0084] Through steps S51 to S53 above, a transparent and secure environment can be provided for multi-party collaboration in life insurance reinsurance business. Private blockchain and node role delineation ensure trusted data sharing, smart contract automatic verification schemes guarantee compliance, and hash value on-chain notarization ensures information is tamper-proof and traceable, thus improving process transparency and security.

[0085] As another optional implementation method, the life insurance reinsurance scheme generation method provided in this application embodiment can also continuously optimize the model and allocation strategy through online learning, specifically including the following steps S61 to S63.

[0086] S61. Obtain feedback data on the life insurance reinsurance allocation plan. The feedback data includes the reinsurance company's actual loss ratio data, interest rate adjustment data, and risk-related data.

[0087] Among them, obtaining feedback data on life insurance reinsurance allocation plans can capture the actual risk performance in business execution (such as the actual loss ratio of reinsurance companies) and changes in the external environment (such as interest rate adjustments) in real time, providing a dynamic basis for model optimization based on real business scenarios and avoiding evaluation bias caused by the model relying on outdated data.

[0088] S62. Set the feedback data as new training data, use the incremental learning mechanism to iteratively optimize the dynamic risk assessment model of life insurance reinsurance, and retain the old value weights of the selected parameters in the dynamic risk assessment model of life insurance reinsurance through the elastic weight consolidation algorithm.

[0089] By using feedback data as new training data and employing an incremental learning mechanism, combined with an elastic weight consolidation algorithm to retain the old values ​​of key parameters, the model can prevent "catastrophic forgetting" of important risk patterns learned in the past (such as long-term loss ratio trends) when introducing new data to update the model. This ensures that the model absorbs new information while maintaining the stability of its core evaluation capabilities during iteration.

[0090] S63. Update the parameters of the life insurance reinsurance dynamic risk assessment model based on the newly added training data to continuously optimize the life insurance reinsurance dynamic risk assessment model.

[0091] Among them, continuously updating model parameters based on new data enables the life insurance reinsurance dynamic risk assessment model to continuously adapt to business changes (such as emerging risk types and market rule adjustments), gradually improving the timeliness and accuracy of risk assessment, providing more realistic risk prediction support for subsequent reinsurance scheme optimization, and forming a closed loop of "data feedback - model iteration - scheme optimization".

[0092] Through steps S61 to S63 above, continuous optimization of the life insurance reinsurance model and strategy can be achieved. Feedback data provides dynamic evidence, incremental learning combined with elastic weights consolidates and avoids forgetting historical patterns, and parameter updates allow the model to adapt to new changes, forming a virtuous cycle and ensuring that the solution meets actual needs.

[0093] The life insurance reinsurance scheme generation method provided in this application achieves precision and intelligence in reinsurance scheme generation through a full-process architecture of multi-source data fusion, dynamic risk assessment, intelligent optimization decision-making, security collaboration, and continuous iteration. It integrates multiple types of data and breaks down data silos through multi-dimensional feature extraction, constructs an evaluation model that integrates temporal convolutional networks and dynamic attention mechanisms, and enhances risk capture capabilities through multi-task training; it generates schemes that balance cost and risk diversification through multi-objective optimization and reinforcement learning, responds to changes through real-time mechanisms; it introduces blockchain to ensure data security, and forms an iterative closed loop through online learning, comprehensively improving reinsurance risk management and operational efficiency, and providing a scientific and efficient solution.

[0094] Please continue reading. Figure 5 , Figure 5 This is a schematic diagram of the system structure of the life insurance reinsurance scheme generation device provided in the embodiments of this application, as shown below. Figure 5As shown, the life insurance reinsurance scheme generation device 50 includes: a business data acquisition module 51, a data cleaning and processing module 52, a risk assessment result module 53, and a life insurance reinsurance scheme module 54.

[0095] The business data acquisition module 51 is specifically used to acquire structured data, unstructured data, and real-time data related to the target life insurance reinsurance business from at least two preset data sources, as multi-source life insurance reinsurance data.

[0096] The data cleaning and processing module 52 is specifically used to clean the multi-source life insurance reinsurance data. It extracts and fuses multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector.

[0097] The risk assessment result module 53 is specifically used to construct a life insurance reinsurance dynamic risk assessment model containing a temporal convolutional network and a dynamic attention mechanism based on the life insurance reinsurance risk feature vector, and output the life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model.

[0098] The life insurance reinsurance scheme module 54 is specifically used to generate a life insurance reinsurance allocation scheme based on the life insurance reinsurance risk assessment results, with the risk-bearing ratio of each reinsurance company as the decision variable, through multi-objective optimization algorithm and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism.

[0099] As an optional implementation, the data cleaning and processing module 52 is further specifically used to extract text feature vectors from the text data in the multi-source life insurance reinsurance data using a pre-trained bidirectional encoder-representation converter model; extract image feature vectors from the image data in the multi-source life insurance reinsurance data using a residual network; construct a reinsurance company association graph using a graph neural network, aggregate neighborhood information based on the reinsurance company association graph to obtain reinsurance company scoring data; input dynamic data from the multi-source life insurance reinsurance data into a temporal convolutional network to generate time-sensitive risk trend feature data; and align the feature dimensions of the text feature vectors, image feature vectors, reinsurance company scoring data, and risk trend feature data using a cross-attention mechanism, and fuse them to generate a unified life insurance reinsurance risk feature vector.

[0100] As an optional implementation, the risk assessment result module 53 is further specifically used to construct a temporal convolutional network using the life insurance reinsurance risk feature vector as input, and to capture long-term temporal dependencies using multi-layer dilated causal convolution; to set target parameters for the dilated convolutional layers of the temporal convolutional network, wherein the kernel size is set to a preset first number, and the dilation factor is set to grow exponentially; to connect a dynamic attention mechanism after the temporal convolutional network, and to calculate the first attention weight of each time step of the hidden state output by the temporal convolutional network; to perform a linear transformation on the hidden state of the current time step and the historical state matrix through the weight matrix, wherein the historical state matrix is ​​generated by the temporal convolutional network after performing multi-layer dilated causal convolution on the input temporal features; to calculate the similarity score between the transformed hidden state of the current time step and the historical state matrix, and to normalize the similarity score to obtain the second attention weight; and to generate a dynamic risk feature vector by weighted summing of the historical state matrix according to the second attention weight, thereby obtaining the constructed dynamic risk assessment model for life insurance reinsurance.

[0101] As an optional implementation, the risk assessment result module 53 is further specifically used to design a loss function for multi-task joint training. The main task of the loss function is to calculate the first loss of the expected payout probability of an individual in the next preset second number of years through binary classification cross-entropy. The auxiliary task of the loss function is to calculate the second loss of the preset volatility data on the life insurance reinsurance risk through mean squared error loss. Using a course learning training strategy, the life insurance reinsurance dynamic risk assessment model is trained through the multi-source life insurance reinsurance data in descending order of confidence level, thereby obtaining the trained life insurance reinsurance dynamic risk assessment model.

[0102] As an optional implementation, the life insurance reinsurance scheme module 54 is further specifically used to set the risk proportion borne by each reinsurance company as a decision variable; construct an objective function that includes cost minimization and risk diversification maximization; the cost minimization is used to comprehensively calculate the product of the reinsurance premium rate and the corresponding risk proportion of each reinsurance company, the product of the risk penalty term and the corresponding risk value, and then sum them up to obtain the total cost, so as to solve for the minimization of the total cost; the risk diversification maximization is used to maximize the degree of risk diversification by allocating the risk proportion of each reinsurance company; and set constraints, including that the risk proportion borne by a single reinsurance company does not exceed a preset first... A risk ratio threshold is set, and the total risk ratio borne by all reinsurance companies is 1. An improved multi-objective evolutionary algorithm and deep reinforcement learning are used for collaborative optimization. The improved multi-objective evolutionary algorithm randomly generates initial solutions that satisfy the constraints, performs adaptive crossover and mutation operations on the parent solutions, performs non-dominated sorting on the solution set, and retains Pareto front solutions. The deep reinforcement learning adjusts the crossover and mutation parameters or population selection strength parameters of the multi-objective evolutionary algorithm based on the characteristics of the current solution set. A real-time triggering mechanism is established. When the dynamic changes in data in a preset data dimension exceed a preset threshold, a local re-optimization process is triggered, retaining the original solutions with a preset ratio as the initial population to accelerate convergence and generate the life insurance reinsurance allocation scheme.

[0103] As an optional implementation, the life insurance reinsurance scheme generation device 50 further includes a private blockchain module. This module is specifically used to construct a private blockchain, set node roles (with the insurance company acting as the initiating node, the reinsurance company as the verification node, and the regulatory agency as the auditing node), create smart contract logic, and automatically verify whether the life insurance reinsurance allocation scheme meets preset verification rules. If the verification passes, the module uploads the life insurance reinsurance risk assessment result and the hash value of the life insurance reinsurance allocation scheme to the private blockchain for storage.

[0104] As an optional implementation, the life insurance reinsurance scheme generation device 50 further includes a model iteration and optimization module. This module is specifically used to acquire feedback data of the life insurance reinsurance allocation scheme, including the reinsurance company's actual loss ratio data, interest rate adjustment data, and risk-related data; set the feedback data as new training data; use an incremental learning mechanism to iteratively optimize the life insurance reinsurance dynamic risk assessment model; retain the old weights of the selected parameters in the life insurance reinsurance dynamic risk assessment model through an elastic weight consolidation algorithm; and update the parameters of the life insurance reinsurance dynamic risk assessment model based on the new training data to continuously optimize the model.

[0105] It should be noted that the aforementioned life insurance reinsurance scheme generation device can execute the life insurance reinsurance scheme generation method provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the embodiments of the life insurance reinsurance scheme generation device can be found in the life insurance reinsurance scheme generation method provided in the embodiments of this application.

[0106] Figure 6 This is a schematic diagram of the hardware structure of the electronic device for executing the life insurance reinsurance plan generation method provided in the embodiments of this application, such as... Figure 6 As shown, the electronic device 600 includes:

[0107] One or more processors 610 and memory 620, Figure 6 Take the 610 processor as an example.

[0108] The processor 610 and the memory 620 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0109] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the life insurance reinsurance scheme generation method in the embodiments of this application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, thereby implementing the life insurance reinsurance scheme generation method in the above-described method embodiments.

[0110] The memory 620 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the life insurance reinsurance scheme generation device. Furthermore, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory remotely located relative to the processor 610, and these remote memories may be connected to the life insurance reinsurance scheme generation device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0111] The one or more modules are stored in the memory 620. When executed by the one or more processors 610, they perform the life insurance reinsurance scheme generation method in any of the above method embodiments, for example, performing the above-described... Figure 2 Method steps S1 to S4, Figure 3 Method steps S21 to S25, Figure 4 Steps S31 to S36 in the method are implemented. Figure 5 The functions of modules 51-54 in the document.

[0112] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0113] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 6 One of the processors 610 can enable the one or more processors to execute the life insurance reinsurance scheme generation method in any of the above method embodiments, for example, to perform the above-described... Figure 2 Method steps S1 to S4, Figure 3 Method steps S21 to S25, Figure 4 Steps S31 to S36 in the method are implemented. Figure 5 The functions of modules 51-54 in the document.

[0114] This application provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, which, when executed by an electronic device, enable the electronic device to perform the life insurance reinsurance scheme generation method in any of the above method embodiments, for example, to perform the above-described method. Figure 2 Method steps S1 to S4, Figure 3 Method steps S21 to S25, Figure 4 Steps S31 to S36 in the method are implemented. Figure 5 The functions of modules 51-54 in the document.

[0115] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for generating a life insurance reinsurance plan, characterized in that, include: Obtain structured data, unstructured data, and real-time data related to the target life insurance reinsurance business from at least two preset data sources, as multi-source life insurance reinsurance data; The multi-source life insurance reinsurance data is cleaned, and multi-dimensional features are extracted and fused from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector. Based on the life insurance reinsurance risk feature vector, a life insurance reinsurance dynamic risk assessment model containing a temporal convolutional network and a dynamic attention mechanism is constructed, and the life insurance reinsurance risk assessment result is output through the trained life insurance reinsurance dynamic risk assessment model. Based on the life insurance reinsurance risk assessment results, and using the risk-bearing ratio of each reinsurance company as the decision variable, a life insurance reinsurance allocation scheme is generated through multi-objective optimization algorithm and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism.

2. The method for generating a life insurance reinsurance plan according to claim 1, characterized in that, The step of extracting and fusing multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector includes: The text feature vectors of the text data in the multi-source life insurance reinsurance data are extracted by a pre-trained bidirectional encoder-representation converter model. Image feature vectors are extracted from the image data in the multi-source life insurance reinsurance data using a residual network. A graph neural network is used to construct a reinsurance company association graph. Neighborhood information is aggregated based on the reinsurance company association graph to obtain reinsurance company scoring data. Input the dynamic data from the multi-source life insurance reinsurance data into a temporal convolutional network to generate time-sensitive risk trend feature data; The feature dimensions of the text feature vector, image feature vector, reinsurance company rating data, and risk trend feature data are aligned by a cross-attention mechanism and fused to generate a unified life insurance reinsurance risk feature vector.

3. The method for generating a life insurance reinsurance plan according to claim 1, characterized in that, The step of constructing a dynamic risk assessment model for life reinsurance that includes a temporal convolutional network and a dynamic attention mechanism based on the life reinsurance risk feature vector includes: Using the life insurance reinsurance risk feature vector as input, a temporal convolutional network is constructed, and multi-layer dilated causal convolution is used to capture long-term temporal dependencies; Set target parameters for the dilated convolutional layers of the temporal convolutional network, wherein the kernel size is set to a preset first number, and the dilation factor is set to grow exponentially. A dynamic attention mechanism is connected after the temporal convolutional network to calculate the first attention weight of the hidden state output by the temporal convolutional network at each time step. The hidden state and the historical state matrix at the current time step are linearly transformed by a weight matrix. The historical state matrix is ​​generated by a temporal convolutional network after performing multi-layer dilated causal convolution on the input temporal features. Calculate the similarity score between the current time step hidden state and the historical state matrix after transformation, and normalize the similarity score to obtain the second attention weight; The historical state matrix is ​​weighted and summed according to the second attention weight to generate a dynamic risk feature vector, thus obtaining the completed dynamic risk assessment model for life insurance reinsurance.

4. The method for generating a life insurance reinsurance plan according to claim 1, characterized in that, Before the step of outputting the life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model, the method further includes: Design a loss function for multi-task joint training, wherein the main task of the loss function is to calculate the first loss of the expected payout probability of an individual in the next preset second number of years through binary classification cross-entropy, and the auxiliary task of the loss function is to calculate the second loss of the life insurance reinsurance risk of preset fluctuation data through mean square error loss. Using the training strategy learned in the course, the dynamic risk assessment model for life reinsurance is trained by the multi-source life reinsurance data in descending order of confidence level, thus obtaining the trained dynamic risk assessment model for life reinsurance.

5. The method for generating a life insurance reinsurance plan according to claim 1, characterized in that, The steps for generating a life insurance reinsurance allocation plan based on the life insurance reinsurance risk assessment results, using the risk-bearing ratio of each reinsurance company as the decision variable, and through multi-objective optimization algorithms and deep reinforcement learning collaborative optimization combined with a real-time triggering mechanism, include: The risk proportion borne by each reinsurance company is set as a decision variable; Construct an objective function that includes both cost minimization and risk diversification maximization; The cost minimization is used to comprehensively calculate the product of the reinsurance premium rate and the corresponding risk ratio of each reinsurance company, the product of the risk penalty term and the corresponding risk value, and then sum them up to obtain the total cost, so as to solve the problem of minimizing the total cost. The aforementioned risk diversification maximization is used to maximize the overall risk distribution by allocating the risk proportions of each reinsurance company. Set constraints, including that the risk ratio borne by a single reinsurance company does not exceed a preset first risk ratio threshold, and the total risk ratio borne by all reinsurance companies is 1. Optimization is achieved through a combination of an improved multi-objective evolutionary algorithm and deep reinforcement learning. The improved multi-objective evolutionary algorithm randomly generates initial solutions that satisfy the constraints, performs adaptive crossover and mutation operations on the parent solutions, performs non-dominated sorting on the solution set, and retains Pareto front solutions. The deep reinforcement learning adjusts the crossover and mutation parameters or the population selection strength parameters of the multi-objective evolutionary algorithm based on the features of the current solution set. A real-time triggering mechanism is established. When the dynamic changes in the data of the preset data dimension exceed the preset threshold, a local re-optimization process is triggered. The original solution of a preset proportion is retained as the initial population to accelerate convergence and generate the life insurance reinsurance allocation scheme.

6. The method for generating a life insurance reinsurance plan according to claim 1, characterized in that, The method for generating life insurance reinsurance plans also includes: Build a private blockchain and set up node roles, with the insurance company as the initiating node, the reinsurance company as the verification node, and the regulatory agency as the auditing node; Create smart contract logic to automatically verify whether the life insurance reinsurance allocation scheme meets the preset verification rules; If the verification is successful, upload the hash values ​​of the life insurance reinsurance risk assessment results and the life insurance reinsurance allocation plan to the private blockchain for storage.

7. The method for generating a life insurance reinsurance plan according to claim 1, characterized in that, The method for generating life insurance reinsurance plans also includes: Obtain feedback data on the life insurance reinsurance allocation plan, including the reinsurance company's actual loss ratio data, interest rate adjustment data, and risk-related data; The feedback data is set as new training data, and the incremental learning mechanism is used to iteratively optimize the dynamic risk assessment model of life reinsurance. The old value weights of the parameters selected in the dynamic risk assessment model of life reinsurance are retained through the elastic weight consolidation algorithm. The parameters of the life insurance reinsurance dynamic risk assessment model are updated based on the newly added training data to continuously optimize the life insurance reinsurance dynamic risk assessment model.

8. A life insurance reinsurance plan generation device, characterized in that, include: The business data acquisition module is used to acquire structured data, unstructured data, and real-time data related to the target life insurance reinsurance business from at least two preset data sources, as multi-source life insurance reinsurance data; The data cleaning and processing module is used to clean the multi-source life insurance reinsurance data. It extracts and fuses multi-dimensional features from the cleaned multi-source life insurance reinsurance data through a multi-dimensional feature extraction mechanism to obtain a life insurance reinsurance risk feature vector. The risk assessment result module is used to construct a life insurance reinsurance dynamic risk assessment model containing a temporal convolutional network and a dynamic attention mechanism based on the life insurance reinsurance risk feature vector, and output the life insurance reinsurance risk assessment result through the trained life insurance reinsurance dynamic risk assessment model. The life insurance reinsurance scheme module is used to generate a life insurance reinsurance allocation scheme based on the life insurance reinsurance risk assessment results, with the risk-bearing ratio of each reinsurance company as the decision variable, through multi-objective optimization algorithm and deep reinforcement learning collaborative optimization, combined with a real-time triggering mechanism.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the life insurance reinsurance scheme generation method according to any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by an electronic device, cause the electronic device to perform the life insurance reinsurance scheme generation method according to any one of claims 1-7.

Images