Planting industry insurance claim data analysis method and system

Abstract

The invention discloses a planting industry insurance claim data analysis method and system, and the method comprises the steps: obtaining planting farmland disaster remote sensing image data, carrying out the recognition and classification of farmland disaster types through a deep learning algorithm, and obtaining a farmland disaster type distribution diagram; according to the farmland disaster type distribution diagram, constructing a disaster situation assessment index system, determining the weight of each index, calculating a comprehensive disaster situation index, and carrying out quantitative assessment on the farmland disaster situation to obtain a farmland disaster situation quantitative assessment result; historical disaster data and corresponding claim settlement data are acquired, and a claim settlement prediction model is constructed through a feature combination and parameter optimization method; and combining a farmland disaster situation quantitative evaluation result with the claim settlement prediction model, inputting disaster situation parameters and predicting a claim settlement amount for a newly occurring farmland disaster, and if the predicted claim settlement amount exceeds a preset threshold value, triggering a claim settlement flow. According to the invention, intelligent assessment and claim settlement prediction of farmland disasters are realized, and the claim settlement efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information analysis, and in particular to a method and system for analyzing crop insurance claims data. Background Art

[0002] With the change of global climate and the transformation of agricultural production mode, the planting industry is facing more and more natural risks and market risks. In order to protect the interests of farmers and promote the sustainable development of agriculture, planting insurance has gradually become an important part of the modern agricultural risk management system. However, with the rapid development of the agricultural insurance market, the amount of planting insurance claims data has increased dramatically, and traditional manual processing methods and simple data analysis tools can no longer meet the needs of the current market.

[0003] Traditional crop insurance claims data analysis methods mainly rely on manual operations and simple statistical tools, resulting in low analysis efficiency, and the analysis results are easily affected by human factors, making it difficult to ensure accuracy. At the same time, the data involved in crop insurance claims are complex and diverse, including planting area, crop types, growth cycles, yields, losses, etc. The correlations between these factors are complex, and traditional analysis methods are difficult to effectively mine the correlation rules and risk characteristics.

[0004] In addition, since crop insurance claims data often involve a large amount of geospatial information, such as the geographical location, climate conditions, soil type, etc. of the planting area, these factors have a significant impact on the growth and yield of crops and are also an important basis for claims decisions. However, traditional data analysis methods often ignore the analysis of these geospatial information, resulting in incomplete and inaccurate analysis results.

[0005] In order to solve the above problems, some crop insurance claims data analysis methods based on big data and geographic information technology have been proposed. However, these methods still have some problems and challenges in practical application, such as difficulty in data acquisition, slow data processing speed, and difficulty in intuitive display of analysis results. Therefore, how to propose an efficient, accurate, and intuitive crop insurance claims data analysis method and system has become a problem that needs to be solved urgently. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a crop insurance claims data analysis method and system to achieve full process automation from disaster monitoring to compensation review.

[0007] On the one hand, to achieve the above-mentioned purpose, the present invention provides a method for analyzing crop insurance claims data, comprising:

[0008] Acquire remote sensing image data of farmland disasters, identify and classify the types of farmland disasters, and obtain a distribution map of farmland disaster types;

[0009] According to the farmland disaster type distribution map, for different disaster types, a disaster assessment index system is constructed, the weight of each index is determined, and a comprehensive disaster index is calculated to quantitatively assess the farmland disaster and obtain a quantitative assessment result of the farmland disaster;

[0010] Obtain historical disaster data and corresponding claims data, and build a claims prediction model through feature combination and parameter optimization methods;

[0011] The quantitative assessment results of the farmland disaster are combined with the claim prediction model. For newly occurring farmland disasters, disaster parameters are input to predict the claim amount. If it is determined that the predicted claim amount exceeds a preset threshold, the claim process is triggered. The claim process is implemented through an intelligent claims platform, which is built based on farmland disaster assessment and claim prediction results, combined with agricultural insurance business rules and risk control requirements.

[0012] Preferably, obtaining the farmland disaster type distribution map includes:

[0013] A convolutional neural network is used to extract multi-scale features of disaster remote sensing image data, a deep learning model is constructed, the extracted multi-scale features are input into the fully connected layer and classification layer of the deep learning model, and the probability of each pixel belonging to a different disaster type is calculated through the softmax function to obtain a probability distribution map of the disaster type;

[0014] According to the probability distribution map of the disaster types, thresholds are set to identify and classify farmland disasters to obtain a distribution map of farmland disaster types.

[0015] Preferably, the farmland disaster is quantitatively assessed to obtain a quantitative assessment result of the farmland disaster, including:

[0016] According to the farmland disaster type distribution map, obtaining distribution information of different disaster types;

[0017] For each type of disaster, determine the corresponding disaster assessment indicators, build a multi-level assessment indicator system, obtain data for each assessment indicator, and convert the data into a form that can be used for analysis;

[0018] The hierarchical analysis method is used to calculate the weight coefficient of each indicator according to the importance of the indicators;

[0019] The index data and the corresponding weight coefficients are weighted and summed to obtain a comprehensive disaster index for each disaster type. According to the size of the comprehensive disaster index, the farmland disaster conditions of different disaster types are quantitatively evaluated and graded to obtain the quantitative evaluation results of the farmland disaster conditions.

[0020] Preferably, constructing the claims prediction model includes:

[0021] Extracting key indicators reflecting the severity of disasters based on the historical disaster data as input features of the claims prediction model;

[0022] Preprocessing the claim data to extract the actual claim amount as the target output value of the model;

[0023] The input features are combined by using a feature combination method to obtain high-dimensional composite features, and the hyperparameters of the extreme gradient boosting tree model are tuned by using a grid search method;

[0024] The optimized extreme gradient boosting tree model is trained, a nonlinear mapping relationship between disaster assessment indicators and claim amounts is established, and the claim prediction model is obtained.

[0025] Preferably, tuning the hyperparameters of the extreme gradient boosting tree model includes:

[0026] Determine the hyperparameters to be optimized for the extreme gradient boosting tree model and form a hyperparameter search space;

[0027] According to the value range and step size of the hyperparameters, a grid search method is used to generate all possible hyperparameter combinations and construct a parameter grid;

[0028] For each set of hyperparameter configurations, the corresponding extreme gradient boosting tree model is trained and the model performance is evaluated on the validation set. The average performance indicators of the model under different hyperparameter configurations are obtained through cross-validation.

[0029] The performance indicators are sorted and the best hyperparameter combination is determined as the optimal parameter configuration of the extreme gradient boosting tree model. The extreme gradient boosting tree model is retrained using the optimal hyperparameters to obtain the final model after performance tuning.

[0030] Preferably, an incremental learning method is used to update and iteratively optimize the claims prediction model, specifically including:

[0031] S1. Obtain continuously collected disaster data and claims data, pre-process the data, and obtain standardized training data sets and test data sets;

[0032] S2. Based on the preprocessed training data set, the prediction model is trained using an incremental learning algorithm to obtain an initial claims prediction model, and the claims prediction model is updated in real time using an online learning method;

[0033] S3. Use the test data set to evaluate the updated claims prediction model and determine whether the model performance meets the preset requirements by calculating the accuracy and recall indicators;

[0034] S4. If the model performance does not meet the requirements, the model is iteratively optimized using hyperparameter optimization and feature selection methods, and S2-S3 are repeated until the model performance is stable and meets the requirements.

[0035] Preferably, the quantitative assessment result of the farmland disaster is combined with the claim prediction model, and for newly occurring farmland disasters, disaster parameters are input to predict the claim amount, including:

[0036] Obtaining quantitative assessment result data of farmland disaster conditions as input parameters of the claim prediction model, and combining the disaster condition parameters to predict the claim amount of the farmland disaster;

[0037] If the predicted claim amount exceeds the threshold, the claim process for farmland disasters will be automatically triggered.

[0038] Preferably, the claims settlement process includes:

[0039] Obtain insurance information of affected farmers, including insured crop type, insured area and insured amount, and associate it with disaster data;

[0040] Through the associated data, combined with the agricultural insurance compensation rules, the decision tree algorithm is used to calculate the amount of compensation to be paid, generate a claim settlement plan, and evaluate the compliance and risk of the claim settlement plan;

[0041] If the assessment is passed, the claim will be reviewed; if not, the claim will be returned for revision.

[0042] Among them, in the compensation review stage, the claim plan is manually reviewed and the compensation amount is confirmed; after the review is passed, the compensation amount will be paid to the farmers to complete the claim process.

[0043] On the other hand, to achieve the above-mentioned purpose, the present invention also provides a crop insurance claim data analysis system, comprising:

[0044] Farmland disaster identification module, used to extract disaster image features and classify them;

[0045] Disaster assessment module, used to build an indicator system and quantitatively assess the disaster situation;

[0046] Claims prediction module, used to build claims prediction model;

[0047] An automatic claim triggering module, used to predict the claim amount through the claim prediction model and trigger the claim process;

[0048] Model optimization module, used to update and iterate the claims prediction model online;

[0049] Intelligent claims platform module, used to automate the entire claims process.

[0050] Compared with the prior art, the present invention has the following advantages and technical effects:

[0051] The present invention first uses a convolutional neural network to extract multi-scale features of farmland disasters from remote sensing images, combines attention mechanism and transfer learning to identify disaster types, and generates a distribution map of farmland disaster types. Then, the analytic hierarchy process is used to determine the indicator weights, and the comprehensive disaster index is calculated to achieve quantitative evaluation. Then, using historical disaster and claims data, with disaster indicators as input and actual claims amounts as output, the extreme gradient boosting tree model is trained to establish a claims prediction model. The disaster assessment results are combined with the claims prediction model to predict the claims amount for newly occurring disasters, and claims are automatically triggered if the threshold is exceeded. The model is continuously optimized through incremental learning to improve prediction accuracy. Finally, through the intelligent claims platform, the full process automation from disaster monitoring to compensation review is realized. The present invention realizes intelligent assessment and claims prediction of farmland disasters, and improves claims efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0053] Figure 1 A flow chart of a method for analyzing claims data in crop insurance according to an embodiment of the present invention;

[0054] Figure 2 The present invention is a schematic diagram of the structure of a crop insurance claims data analysis system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0057] The present invention proposes a method for analyzing crop insurance claims data. Figure 1 ,include:

[0058] Acquire remote sensing image data of farmland disasters, identify and classify farmland disaster types through deep learning algorithms, and obtain a distribution map of farmland disaster types;

[0059] According to the farmland disaster type distribution map, for different disaster types, a disaster assessment index system is constructed, the weight of each index is determined, and a comprehensive disaster index is calculated to quantitatively assess the farmland disaster and obtain a quantitative assessment result of the farmland disaster;

[0060] Obtain historical disaster data and corresponding claims data, and build a claims prediction model through feature combination and parameter optimization methods;

[0061] Combine the quantitative assessment results of farmland disasters with the claims prediction model, input disaster parameters for newly occurring farmland disasters, and predict the claims amount. If the predicted claims amount exceeds the preset threshold, the claims flow is triggered, wherein the claims process is implemented through an intelligent claims platform, which is built based on farmland disaster assessment and claims prediction results, combined with agricultural insurance business rules and risk control requirements.

[0062] The present invention first uses convolutional neural networks to extract multi-scale features of farmland disasters from remote sensing images, combines attention mechanisms and transfer learning to identify disaster types, and generates a distribution map of farmland disaster types. Then, the analytic hierarchy process is used to determine the indicator weights, and the comprehensive disaster index is calculated to achieve quantitative evaluation. Then, using historical disaster and claims data, with disaster indicators as input and actual claims amounts as output, the extreme gradient boosting tree model is trained to establish a claims prediction model. The disaster assessment results are combined with the claims prediction model to predict the claims amount for newly occurring disasters, and claims are automatically triggered if the threshold is exceeded. The model is continuously optimized through incremental learning to improve prediction accuracy. Finally, through the intelligent claims platform, the entire process from disaster monitoring to compensation review is automated.

[0063] Furthermore, a distribution map of farmland disaster types is obtained, including:

[0064] A convolutional neural network is used to extract multi-scale features of disaster remote sensing image data, a deep learning model is constructed, the extracted multi-scale features are input into the fully connected layer and classification layer of the deep learning model, and the probability of each pixel belonging to a different disaster type is calculated through the softmax function to obtain a probability distribution map of the disaster type;

[0065] According to the probability distribution map of disaster types, thresholds are set to identify and classify farmland disasters, and a distribution map of farmland disaster types is obtained.

[0066] Specifically, remote sensing image data of cultivated farmland disasters is obtained, and the image data is preprocessed, including radiation calibration, atmospheric correction and geometric correction, to obtain standardized remote sensing images. Convolutional neural networks are used to extract features of preprocessed remote sensing images. By setting convolution kernels of different scales and sizes, multi-scale texture, shape and color features of disaster images are extracted to obtain feature representations of disaster images. The attention mechanism is introduced to give different weights to different areas of disaster images, highlight the features of disaster areas, suppress the features of non-disaster areas, and improve the accuracy of disaster identification. Using the transfer learning method, convolutional neural networks are pre-trained on a large number of annotated disaster image data sets to learn the general feature representation of disaster images, and then fine-tuned on the target farmland disaster image data to improve the generalization ability and robustness of the model. A deep learning model is constructed, and the extracted disaster image features are input into the fully connected layer and the classification layer. The probability of each pixel belonging to different disaster types is calculated by the softmax function to obtain the probability distribution map of disaster types. According to the probability distribution map of disaster types, the threshold is set to identify and classify farmland disasters, and the pixels with probability greater than the threshold are divided into corresponding disaster types to obtain the distribution map of farmland disaster types. The farmland disaster type distribution map is post-processed, including removing small areas and smoothing boundaries, to obtain the final farmland disaster type distribution map, which provides a basis for the assessment and decision-making of farmland disasters.

[0067] Furthermore, the farmland disaster is quantitatively assessed to obtain the quantitative assessment results of the farmland disaster, including:

[0068] According to the farmland disaster type distribution map, obtain the distribution information of different disaster types;

[0069] For each type of disaster, determine the corresponding disaster assessment indicators, build a multi-level assessment indicator system, obtain data for each assessment indicator, and convert the data into a form that can be used for analysis;

[0070] The hierarchical analysis method is used to calculate the weight coefficient of each indicator according to the importance of the indicators;

[0071] The index data and the corresponding weight coefficients are weighted and summed to obtain a comprehensive disaster index for each disaster type. According to the size of the comprehensive disaster index, the farmland disaster conditions of different disaster types are quantitatively evaluated and graded to obtain the quantitative evaluation results of the farmland disaster conditions.

[0072] Specifically, according to the distribution map of farmland disaster types, the distribution information of different disaster types is obtained; for each disaster type, the corresponding disaster assessment indicators are determined, and a multi-level assessment indicator system is constructed; the data of each assessment indicator, including the affected area, the degree of disaster, etc., are obtained, and the data are converted into a form that can be used for analysis through data preprocessing and feature extraction; the hierarchical analysis method is used to calculate the weight coefficient of each indicator according to the degree of importance between the indicators; the preprocessed indicator data and the corresponding weight coefficient are weighted and summed to obtain the comprehensive disaster index of each disaster type; according to the size of the comprehensive disaster index, the farmland disasters of different disaster types are quantitatively evaluated and graded; the assessment results and the farmland disaster distribution map are visualized to intuitively present the spatial distribution characteristics of the disaster and provide decision support for post-disaster rescue and reconstruction.

[0073] Furthermore, a claims prediction model is constructed, including:

[0074] Based on historical disaster data, key indicators reflecting the severity of the disaster are extracted as input features of the model;

[0075] Preprocess the claims data and extract the actual claims amount as the target output value of the model;

[0076] The feature combination method is used to combine the input indicators to obtain high-dimensional composite features, and the hyperparameters of the extreme gradient boosting tree model are tuned through the grid search method;

[0077] The optimized extreme gradient boosting tree model is trained to establish a nonlinear mapping relationship between disaster assessment indicators and claim amounts, and obtain a claim prediction model.

[0078] Specifically, historical disaster data containing information such as the time, location, type, affected area, and degree of disaster, as well as the corresponding insurance claims record data, are obtained. Based on the historical disaster data, key indicators that can reflect the severity of the disaster, such as rainfall, wind speed, magnitude, and flood depth, are extracted as input features of the model. The claims data is preprocessed and the actual claim amount is extracted as the target output value of the model. The feature combination method is used to combine multiple input indicators into composite features with higher dimensions and more information to improve the fitting ability of the model. The hyperparameters of the extreme gradient boosting tree model are tuned by parameter optimization methods such as grid search, such as the number of trees, maximum depth, and learning rate, to improve the prediction accuracy of the model. The optimized extreme gradient boosting tree model is trained to establish a nonlinear mapping relationship between the disaster assessment index and the claim amount, and a claim prediction model is constructed. When a new disaster event occurs, the corresponding evaluation index is extracted according to the disaster information and input into the claim prediction model, so that the claim amount of the disaster can be predicted, providing a quantitative reference for the insurance company's claim decision.

[0079] Furthermore, the hyperparameters of the extreme gradient boosting tree model are tuned, including:

[0080] Determine the hyperparameters to be optimized for the extreme gradient boosting tree model and form a hyperparameter search space;

[0081] According to the value range and step size of the hyperparameters, a grid search method is used to generate all possible hyperparameter combinations and construct a parameter grid;

[0082] For each set of hyperparameter configurations, the corresponding extreme gradient boosting tree model is trained and the model performance is evaluated on the validation set. The average performance indicators of the model under different hyperparameter configurations are obtained through cross-validation.

[0083] The performance indicators are sorted and the best hyperparameter combination is determined as the optimal parameter configuration of the extreme gradient boosting tree model. The extreme gradient boosting tree model is retrained using the optimal hyperparameters to obtain the final model after performance tuning.

[0084] Specifically, the hyperparameters to be optimized of the extreme gradient boosting tree model, including the number of trees, maximum depth, and learning rate, are determined to form a hyperparameter search space. According to the value range and step size of the hyperparameters, a grid search method is used to generate all possible hyperparameter combinations and construct a parameter grid. For each set of hyperparameter configurations, the corresponding extreme gradient boosting tree model is trained, and the model performance is evaluated on the validation set. Through cross-validation, the average performance indicators of the model under different hyperparameter configurations, such as accuracy and F1 value, are obtained. According to the ranking of performance indicators, the best performing hyperparameter combination is determined as the optimal parameter configuration of the extreme gradient boosting tree model. The extreme gradient boosting tree model is retrained with the optimal hyperparameters to obtain the final model after performance tuning. The tuned extreme gradient boosting tree model is applied to the actual prediction task, and its generalization performance on the test set is evaluated to verify the effectiveness of parameter optimization.

[0085] Furthermore, the incremental learning method is used to update and iteratively optimize the claims prediction model, including:

[0086] S1. Obtain continuously collected disaster data and claims data, pre-process the data, and obtain standardized training data sets and test data sets;

[0087] S2. Based on the preprocessed training data set, the prediction model is trained using an incremental learning algorithm to obtain an initial claims prediction model, and the claims prediction model is updated in real time using an online learning method;

[0088] S3. Use the test data set to evaluate the updated claims prediction model and determine whether the model performance meets the preset requirements by calculating the accuracy and recall indicators;

[0089] S4. If the model performance does not meet the requirements, the model is iteratively optimized using hyperparameter optimization and feature selection methods, and S2-S3 are repeated until the model performance is stable and meets the requirements.

[0090] Specifically, the continuously collected disaster data and claims data are obtained, and the data are preprocessed, including data cleaning and feature extraction, to obtain standardized training data sets and test data sets. According to the preprocessed training data set, the prediction model is trained using an incremental learning algorithm to obtain an initial claims prediction model. For the newly collected new disaster data and claims data, the online learning method is used to update the claims prediction model in real time so that the model can adapt to the new data distribution and change trend. In the process of model updating, the robustness and generalization ability of the model are improved and the risk of overfitting is reduced by introducing technologies such as regularization terms and ensemble learning. The updated claims prediction model is regularly evaluated using the test data set, and the model performance is judged to meet the preset threshold requirements by calculating indicators such as accuracy and recall rate. If the model performance does not meet the standard, the model is iteratively optimized using methods such as hyperparameter optimization and feature selection until the model performance is stable and meets the requirements. The optimized claims prediction model is used to predict claims for the newly collected disaster data, and the prediction results are output to provide a reference for the insurance company's claims decision-making and improve the efficiency and accuracy of claims processing.

[0091] Furthermore, the quantitative assessment results of farmland disasters are combined with the claims prediction model, and for newly occurring farmland disasters, disaster parameters are input to predict the claims amount, including:

[0092] Obtaining quantitative assessment result data of farmland disaster conditions as input parameters of the claim prediction model, and combining the disaster condition parameters to predict the claim amount of the farmland disaster;

[0093] Determine whether the predicted claim amount exceeds a preset claim amount threshold;

[0094] If the predicted claim amount exceeds the threshold, the claim process for farmland disasters will be automatically triggered.

[0095] Specifically, farmland disaster data, including disaster type, occurrence time, affected area and other information, are obtained and stored in the disaster database. The loss degree of the affected farmland is evaluated to obtain the estimated loss amount. The estimated loss amount is compared with the agricultural insurance business rules to determine whether it meets the claim standard. If the claim standard is met, the claim process is triggered; if not, the process is exited. In the claim process, the insurance information of the affected farmers, including the insured crop type, insured area, and insured amount, is obtained and associated with the disaster data. Through the associated data, combined with the agricultural insurance payment rules, the decision tree algorithm is used to calculate the amount to be paid and generate a claim plan. The compliance and risk of the claim plan are evaluated. If the evaluation is passed, it will enter the payment review stage; if not, it will be returned for modification. In the payment review stage, the claim plan is manually reviewed and the payment amount is confirmed. After the review is passed, the compensation amount is paid to the farmers to complete the claim process. At the same time, the claim data is returned to the system to optimize the loss assessment and risk control models and improve the intelligent level of claims.

[0096] Furthermore, the claims process includes:

[0097] Obtain insurance information of affected farmers, including insured crop type, insured area and insured amount, and associate it with disaster data;

[0098] Through the associated data, combined with the agricultural insurance compensation rules, the decision tree algorithm is used to calculate the amount of compensation to be paid, generate a claim settlement plan, and evaluate the compliance and risk of the claim settlement plan;

[0099] If the assessment is passed, the claim will be reviewed; if not, the claim will be returned for revision.

[0100] Among them, in the compensation review stage, the claim plan is manually reviewed and the compensation amount is confirmed; after the review is passed, the compensation amount will be paid to the farmers to complete the claim process.

[0101] Preferably, the estimated loss amount is compared with the deductible and total insured amount in the agricultural insurance terms, and the decision tree model is used to determine whether the claim standard is met. If the claim standard is met, the claim service is called through the RPC interface to pass in the insurance information of the affected farmers. The claim service uses the decision tree algorithm to calculate the amount to be paid based on the agricultural insurance payment rules, combined with factors such as the estimated loss amount and the insured area, and generates a claim plan. The claim plan is automatically reviewed by a rule-based expert system in the form of a Kafka message queue. The approved claim plan is transferred to the payment link, and the RPA robot automatically initiates a payment request. After manual review and confirmation, the compensation amount is transferred to the farmer's account through the bank-enterprise direct connection. The claim data of each paid case is returned to the data warehouse through the ETL tool for optimizing the model and rules.

[0102] This embodiment also provides a crop insurance claim data analysis system, such as Figure 2 ,include:

[0103] Farmland disaster identification module, used to extract disaster image features and classify them;

[0104] Disaster assessment module, used to build an indicator system and quantitatively assess the disaster situation;

[0105] Claims prediction module, used to build claims prediction model;

[0106] An automatic claim triggering module, used to predict the claim amount through the claim prediction model and trigger the claim process;

[0107] Model optimization module, used to update and iterate the claims prediction model online;

[0108] Intelligent claims platform module, used to automate the entire claims process.

[0109] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for analyzing crop insurance claims data, characterized in that: include: Acquire remote sensing image data of farmland disasters, identify and classify the types of farmland disasters, and obtain a distribution map of farmland disaster types; According to the farmland disaster type distribution map, for different disaster types, a disaster assessment index system is constructed, the weight of each index is determined, and a comprehensive disaster index is calculated to quantitatively assess the farmland disaster and obtain a quantitative assessment result of the farmland disaster; Obtain historical disaster data and corresponding claims data, and build a claims prediction model through feature combination and parameter optimization methods; The quantitative assessment results of the farmland disaster are combined with the claim prediction model. For newly occurring farmland disasters, disaster parameters are input to predict the claim amount. If it is determined that the predicted claim amount exceeds a preset threshold, the claim process is triggered. The claim process is implemented through an intelligent claims platform, which is built based on farmland disaster assessment and claim prediction results, combined with agricultural insurance business rules and risk control requirements.

2. The method for analyzing crop insurance claims data according to claim 1, characterized in that: Obtaining the distribution map of farmland disaster types, including: A convolutional neural network is used to extract multi-scale features of disaster remote sensing image data, a deep learning model is constructed, the extracted multi-scale features are input into the fully connected layer and classification layer of the deep learning model, and the probability of each pixel belonging to a different disaster type is calculated through the softmax function to obtain a probability distribution map of the disaster type; According to the probability distribution map of the disaster types, thresholds are set to identify and classify farmland disasters to obtain a distribution map of farmland disaster types.

3. The method for analyzing crop insurance claims data according to claim 1, characterized in that: Conduct a quantitative assessment of the farmland disaster to obtain a quantitative assessment result of the farmland disaster, including: According to the farmland disaster type distribution map, obtaining distribution information of different disaster types; For each type of disaster, determine the corresponding disaster assessment indicators, build a multi-level assessment indicator system, obtain data for each assessment indicator, and convert the data into a form that can be used for analysis; The hierarchical analysis method is used to calculate the weight coefficient of each indicator according to the importance of the indicators; The index data and the corresponding weight coefficients are weighted and summed to obtain a comprehensive disaster index for each disaster type. According to the size of the comprehensive disaster index, the farmland disaster conditions of different disaster types are quantitatively evaluated and graded to obtain the quantitative evaluation results of the farmland disaster conditions.

4. The method for analyzing crop insurance claims data according to claim 1, characterized in that: Constructing the claims prediction model includes: Extracting key indicators reflecting the severity of disasters based on the historical disaster data as input features of the claims prediction model; Preprocessing the claim data to extract the actual claim amount as the target output value of the model; The input features are combined by using a feature combination method to obtain high-dimensional composite features, and the hyperparameters of the extreme gradient boosting tree model are tuned by using a grid search method; The optimized extreme gradient boosting tree model is trained, a nonlinear mapping relationship between disaster assessment indicators and claim amounts is established, and the claim prediction model is obtained.

5. The method for analyzing crop insurance claims data according to claim 4, characterized in that: The hyperparameters of the extreme gradient boosting tree model are tuned, including: Determine the hyperparameters to be optimized for the extreme gradient boosting tree model and form a hyperparameter search space; According to the value range and step size of the hyperparameters, a grid search method is used to generate all possible hyperparameter combinations and construct a parameter grid; For each set of hyperparameter configurations, the corresponding extreme gradient boosting tree model is trained and the model performance is evaluated on the validation set. The average performance indicators of the model under different hyperparameter configurations are obtained through cross-validation. The performance indicators are sorted and the best hyperparameter combination is determined as the optimal parameter configuration of the extreme gradient boosting tree model. The extreme gradient boosting tree model is retrained using the optimal hyperparameters to obtain the final model after performance tuning.

6. The method for analyzing crop insurance claims data according to claim 5, characterized in that: The incremental learning method is used to update and iteratively optimize the claims prediction model, specifically including: S1. Obtain continuously collected disaster data and claims data, pre-process the data, and obtain standardized training data sets and test data sets; S2. Based on the preprocessed training data set, the prediction model is trained using an incremental learning algorithm to obtain an initial claims prediction model, and the claims prediction model is updated in real time using an online learning method; S3. Use the test data set to evaluate the updated claims prediction model and determine whether the model performance meets the preset requirements by calculating the accuracy and recall indicators; S4. If the model performance does not meet the requirements, the model is iteratively optimized using hyperparameter optimization and feature selection methods, and S2-S3 are repeated until the model performance is stable and meets the requirements.

7. The method for analyzing crop insurance claims data according to claim 1, characterized in that: The method combines the quantitative assessment results of farmland disaster with the claim prediction model, inputs disaster parameters for newly occurring farmland disasters, and predicts the claim amount, including: Obtaining quantitative assessment result data of farmland disaster conditions as input parameters of the claim prediction model, and combining the disaster condition parameters to predict the claim amount of the farmland disaster; If the predicted claim amount exceeds the threshold, the claim process for farmland disasters will be automatically triggered.

8. The method for analyzing crop insurance claims data according to claim 1, characterized in that: The claims process includes: Obtain insurance information of affected farmers, including insured crop type, insured area and insured amount, and associate it with disaster data; Through the associated data, combined with the agricultural insurance compensation rules, the decision tree algorithm is used to calculate the amount of compensation to be paid, generate a claim settlement plan, and evaluate the compliance and risk of the claim settlement plan; If the assessment is passed, the claim will be reviewed; if not, the claim will be returned for revision. Among them, in the compensation review stage, the claim plan is manually reviewed and the compensation amount is confirmed; after the review is passed, the compensation amount will be paid to the farmers to complete the claim process.

9. A crop insurance claim data analysis system, applied to the crop insurance claim data analysis method according to any one of claims 1 to 8, characterized in that: include: Farmland disaster identification module, used to extract disaster image features and classify them; Disaster assessment module, used to build an indicator system and quantitatively assess the disaster situation; Claims prediction module, used to build claims prediction model; An automatic claim triggering module, used to predict the claim amount through the claim prediction model and trigger the claim process; Model optimization module, used to update and iterate the claims prediction model online; Intelligent claims platform module, used to automate the entire claims process.

Images