Machine learning based cargo transportation insurance matching method, system, and medium

Abstract

The application discloses a cargo transportation insurance matching method and system based on machine learning and a medium, and belongs to the field of cargo transportation. The method comprises the following steps: reading transportation data of a target cargo transportation task, wherein the transportation data comprises cargo attributes, a transportation route, a carrier subject and historical transportation claim data; analyzing the cargo attribute data to establish cargo transportation demand characteristics; executing carrier subject data analysis to establish carrier capacity characteristics; performing deviation analysis based on the cargo transportation demand characteristics and the carrier capacity characteristics to establish transportation mismatch degree characteristics; performing risk correlation analysis using the transportation route data to establish transportation risk characteristics; performing similar scenario matching based on the risk characteristics and the transportation claim data, using the matching result to perform matching, and outputting the matching result. The application solves the technical problem that the existing cargo transportation insurance matching cannot realize dynamic matching of cargo transportation risks and insurance schemes, resulting in resource waste caused by mismatch between insurance protection and actual risks of cargo transportation.

Description

Technical Field

[0001] This invention relates to the field of cargo transportation, and more specifically to a cargo transportation insurance matching method, system, and medium based on machine learning. Background Technology

[0002] Currently, the cargo transportation insurance industry still generally adopts a standardized insurance matching model based on manual experience. This model fails to quantify the multi-dimensional mismatch between the differentiated transportation needs of different goods and the actual transportation capacity of carriers, and it also fails to establish a causal transmission relationship between the degree of transportation mismatch and transportation routes. It ignores the spatiotemporal accumulation and amplification effect of mismatch risks in different transportation segments. At the same time, the insurance matching process does not fully incorporate real payout data and actual loss patterns from historical claims cases, and only makes recommendations based on the theoretical coverage of insurance clauses. As a result, it is impossible to quantify the risks of the entire cargo transportation chain and dynamically match insurance plans. This leads to a serious mismatch between insurance coverage and the actual risks of cargo transportation. This results in frequent gaps in cargo damage protection caused by uncovered risks, and there is also a widespread waste of premium resources due to excessive insurance coverage for unnecessary risks. It is difficult to meet the multi-category and refined insurance protection needs of the modern logistics industry. Summary of the Invention

[0003] This application provides a machine learning-based method, system, and medium for matching cargo transportation insurance, aiming to solve the technical problem that existing cargo transportation insurance matching methods are unable to achieve dynamic matching between cargo transportation risks and insurance schemes, leading to a mismatch between insurance coverage and actual cargo transportation risks and resulting in resource waste. The goal is to improve the efficiency of cargo transportation insurance coverage and the utilization rate of resources.

[0004] In view of the above problems, this application provides a method, system and medium for matching cargo transportation insurance based on machine learning.

[0005] Firstly, a machine learning-based method for matching cargo transportation insurance is provided, which includes: The system reads transportation data for the target cargo transportation task, including cargo attribute data, transportation route data, carrier data, and historical transportation claims data. It then performs transportation demand analysis on the cargo attribute data to establish cargo transportation demand characteristics that characterize the target cargo's requirements for transportation conditions during the transportation process. Next, it performs transportation capacity analysis on the carrier data, using historical transportation records, transportation equipment information, and transportation performance data to establish carrier capacity characteristics that characterize the carrier's actual transportation capacity. Based on the cargo transportation demand characteristics and carrier capacity characteristics, it conducts adaptation deviation analysis to establish transportation mismatch characteristics. Finally, it performs historical risk event correlation analysis on the transportation segments corresponding to the target transportation route using transportation route data to establish the risk amplification trend of transportation mismatch characteristics during the transportation process, thus establishing transportation risk characteristics. Finally, it performs similar scenario matching based on the transportation risk characteristics and historical transportation claims data, uses the similar scenario matching results to perform insurance coverage matching, and outputs insurance matching results.

[0006] Secondly, a machine learning-based cargo transportation insurance matching system is provided, which includes: The system comprises the following modules: a reading module for reading transportation data for a target cargo transportation task, including cargo attribute data, transportation route data, carrier data, and historical transportation claims data; a parsing module for parsing the cargo attribute data to analyze transportation demand and establish cargo transportation demand characteristics that characterize the target cargo's requirements for transportation conditions during transportation; an execution module for analyzing the carrier data to analyze its transportation capacity and establishing carrier capacity characteristics that characterize the carrier's actual transportation capacity using historical transportation records, transportation equipment information, and transportation performance data; an analysis module for performing adaptation deviation analysis based on cargo transportation demand characteristics and carrier capacity characteristics to establish transportation mismatch characteristics; a risk module for performing historical risk event correlation analysis on the transportation segment corresponding to the target transportation route using transportation route data, establishing the risk amplification trend of transportation mismatch characteristics during transportation, and establishing transportation risk characteristics; and a matching module for performing similar scenario matching based on transportation risk characteristics and historical transportation claims data, using the similar scenario matching results to perform insurance coverage matching, and outputting insurance matching results.

[0007] Thirdly, this application provides a computer-readable storage medium storing a computer program for executing the machine learning-based cargo transportation insurance matching method provided in this application.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: By structurally analyzing and feature-engineering multi-source heterogeneous transportation data, a dimensional adaptation deviation quantification system for cargo transportation demand and carrying capacity is established. A spatiotemporal risk amplification inference network based on a structural causal model is constructed to accurately depict the evolution of risk along the transportation path. Combined with weighted cosine similarity, accurate retrieval of similar claims scenarios and quantitative assessment of the actual coverage of insurance types are achieved. Finally, a personalized insurance combination scheme that highly matches the actual transportation risks is output, effectively reducing the cost of cargo transportation insurance and improving the protection efficiency and resource utilization of cargo transportation insurance.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a machine learning-based cargo transportation insurance matching method is provided for embodiments of this application. Figure 2 A schematic diagram of the structure of a machine learning-based cargo transportation insurance matching system is provided for embodiments of this application; Figure 3 The present application provides a risk time-series evolution curve for a machine learning-based cargo transportation insurance matching method.

[0011] Explanation of reference numerals in the attached diagram: Reading module 11, Parsing module 12, Execution module 13, Analysis module 14, Risk module 15, Matching module 16. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] The overall concept of the technical solution provided in this application is as follows: This application provides a machine learning-based method, system, and medium for matching cargo transportation insurance. By reading transportation data of the target cargo transportation task, including cargo attributes, transportation routes, carriers, and historical transportation claims data; parsing the cargo attribute data to establish cargo transportation demand characteristics; parsing the carrier data to establish carrier capacity characteristics; performing deviation analysis based on the cargo transportation demand characteristics and carrier capacity characteristics to establish transportation mismatch characteristics; using transportation route data to perform risk correlation analysis to establish transportation risk characteristics; matching similar scenarios based on risk characteristics and transportation claims data; and using the matching results to perform matching, ultimately outputting an insurance combination scheme that highly matches the actual risk, effectively reducing cargo transportation insurance costs and improving the protection efficiency and resource utilization of cargo transportation insurance.

[0014] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0015] Example 1, as Figure 1 As shown in the embodiments of this application, a cargo transportation insurance matching method based on machine learning is provided, the method including: S100: Read the transportation data of the target cargo transportation task, including cargo attribute data, transportation route data, carrier data and historical transportation claims data.

[0016] Specifically, a two-way data channel is first established with the cargo owner's system, order management system, logistics company transportation management system, geographic information system, Ministry of Transport's supervision platform, and insurance company's business system. Upon receiving the trigger instruction for the target cargo transportation task, the system automatically pulls four types of transportation data from the above systems and completes preliminary cleaning.

[0017] Among them, cargo attribute data is a dataset describing the inherent physical, chemical, and economic attributes of the cargo itself. In addition to directly extracting structured fields from electronic waybills and customs declarations, the system also uses OCR technology to recognize unstructured cargo manuals and packaging labels to obtain basic information such as cargo category, unit weight and volume, total value, and packaging type and material. Simultaneously, it parses the impact fragility value (the maximum impact acceleration the cargo can withstand; exceeding this value will cause irreversible damage) and the temperature tolerance gradient (the rate at which the cargo's performance deteriorates per unit temperature change, used to quantify the cargo's sensitivity to temperature fluctuations). Value density, or the economic value of goods per unit volume or weight, directly determines key vulnerability parameters such as the scale of compensation for cargo damage accidents, as well as special category identifiers such as fragile, flammable, explosive, and perishable goods. Transportation route data is a dataset recording the complete transportation path of goods from origin to destination, along with environmental and infrastructure information for each segment. The system first generates the optimal transportation route based on the origin, destination, and transit nodes in the electronic waybill, then breaks it down into N consecutive transportation segments—independent segment units divided according to administrative regions, road types, or terrain features. Subsequently, the system retrieves the terrain ruggedness of each segment from the GIS platform and road network monitoring platform. The degree of surface undulation of a road section calculated by a digital elevation model directly affects the vibration amplitude and stability of vehicles, the probability of seasonal strong winds or rainfall (based on the frequency of extreme weather events from the same period over the past five years), and is a key indicator for predicting cargo damage caused by natural disasters. Road surface grade, defined according to highway engineering technical standards (e.g., expressways, gravel roads), determines the degree of vehicle bumps and the probability of accidents. Additional data include traffic congestion, bridge height and weight restrictions, and toll station travel time. The carrier data reflects the comprehensive qualifications and operational capabilities of the logistics companies, vehicles, and drivers undertaking the transportation task. The system synchronizes the business licenses, road transport operation permits, and safety transport ratings for the past three years of the carrier companies from the logistics system and the road transport supervision platform. It also extracts historical transport records, namely the quantity and performance of similar cargo and route transport tasks completed by the carrier companies in the past, transport equipment information, namely the vehicle type, age, load capacity, and whether it is equipped with hardware parameters such as temperature control system, shock absorption device, and GPS positioning device, and transport performance data, namely the on-time rate, cargo damage rate, cargo loss rate, customer complaint rate and other core operational indicators of historical transport tasks, and links them to the driver's driving experience, safe driving mileage, violation records and professional qualification certificate information for this task.Historical transportation claims data is a complete record of past cargo transportation insurance claims accumulated by insurance companies. It includes information on the entire process of accident occurrence, loss assessment, and compensation. The system retrieves data from the insurance company's claims management system for all cases within the past five years involving the same product category, route, and carrier type as this mission. This data covers the time and location of the accident, the transportation segment involved, the cause of the accident, the type of loss, the amount of loss, the amount of compensation, the type of insurance, and the claims processing time. Simultaneously, the system extracts the original transportation data corresponding to each case.

[0018] Finally, after completing the multi-source data retrieval, the system will automatically perform data integrity verification and outlier detection, remove invalid data with missing value, origin and other key fields, correct logical errors such as negative transportation time and value exceeding reasonable range, and convert all unstructured data into a unified structured format before storing it in a distributed data warehouse.

[0019] S200: Perform transportation demand analysis on the cargo attribute data in the transportation data to establish cargo transportation demand characteristics that characterize the transportation conditions required for the target cargo during transportation.

[0020] Specifically, the system first calls a pre-trained deep learning model for cargo category-demand mapping to perform fine-grained three-level classification of cargo. General categories, including electronic products, fresh food, and chemical raw materials, are further broken down into atomic subcategories with clear transportation demand characteristics, such as laptops with screens in consumer electronics and high-precision accelerometers in industrial electronics. Then, core vulnerability parameters such as impact fragility, temperature change tolerance gradient, and value density are extracted from cargo attribute data. Combined with the cargo's packaging type, unit weight, total value, and special safety markings, five core protection dimensions that completely correspond to the subsequent adaptation deviation analysis are precisely quantified and analyzed.

[0021] In terms of temperature control, the system calculates the total allowable cumulative temperature fluctuation of the goods during transportation based on the temperature tolerance gradient of the goods, i.e. the rate at which the performance of the goods deteriorates under a unit temperature change (the larger the value, the more sensitive the goods are to temperature fluctuations) and the expected transportation time. This is done through a heat conduction model to determine the precise temperature control range, i.e. the upper and lower limits of the temperature that must be maintained throughout the transportation process, including the 2-8℃ requirement for cold chain vaccines and the -18℃ or lower requirement for frozen seafood. The system also indicates whether constant temperature control is required throughout the process, whether a short-term temperature exceedance of no more than 30 minutes is allowed, and the emergency handling requirements after the exceedance.

[0022] In terms of vibration control, the system considers the impact fragility value of the goods—the maximum impact acceleration that the goods can withstand, exceeding which irreversible structural damage will occur—and the cushioning performance parameters of packaging materials, including the elastic modulus of foam and the compressive strength of corrugated paper, to calculate the maximum vibration acceleration threshold that the goods can withstand. It then generates corresponding vibration power spectral density requirements, parameters describing the permissible vibration energy distribution at different frequencies, thus clarifying differentiated vibration control standards for different modes of transportation such as road, rail, and air. In terms of delivery timeliness, the system analyzes the maximum permissible delay time based on the goods' usage scenario, including emergency relief supplies, pre-sale goods during e-commerce promotions, and shelf-life or expiration date requirements. This means that if the delivery time exceeds the agreed time, it will result in the complete loss or loss of the goods' value. The system determines the critical timeframe for partial loss and classifies goods into four levels of time sensitivity: Level 1, Level 2, and Level 3. For Level 1 time-sensitive goods, such as cardiac stents and emergency medications, the maximum allowable delay is typically no more than 2 hours. In terms of loading and unloading operations, the system uses OCR to identify markings on the packaging indicating fragility, prohibition of inversion, and lifting point locations. Combined with the weight of individual items, center of gravity offset, and stacking strength, the system generates standardized loading and unloading procedures and quantifies them into requirements for operational standardization. This includes quantitative scoring standards for operational actions, tool usage, and stacking methods during loading and unloading, such as prohibiting throwing, ensuring that the weight of a single item handled manually does not exceed 25 kg, and limiting the number of stacked layers to no more than 3. The system also clarifies whether specialized forklifts, cranes, and other loading and unloading equipment, as well as certified operators, are required.

[0023] In terms of packaging impact resistance, the system calculates the packaging impact resistance margin based on the packaging material, thickness, and cushioning structure design, including honeycomb structures, air column bags, and other sealing methods. This margin is the difference between the maximum impact protection capability that the packaging can provide and the impact fragility value of the goods. The smaller the margin, the higher the probability of the goods being damaged by impact. The system also indicates whether the packaging needs additional stretch film reinforcement and whether it is suitable for mixing with heavy goods.

[0024] Based on this, the system will perform min-max standardization on the quantification results of the five dimensions, convert them into numerical features in the 0-1 range, and then supplement them with binarized features for special requirements such as explosion and leakage prevention of dangerous goods, full-process GPS monitoring of valuables, and ventilation requirements for perishable goods. Finally, the system will splice them together to form a cargo transportation demand feature vector with unified dimensions that can be directly used in subsequent calculations. Each component of this vector corresponds precisely to a quantifiable and verifiable transportation condition requirement.

[0025] S300: Performs transport capacity analysis on carrier data, and establishes transport capacity characteristics that characterize the actual transport capacity of the carrier by using historical transport records, transport equipment information, and transport performance data.

[0026] Specifically, the system first establishes five capability assessment dimensions that correspond one-to-one with cargo transportation needs. These dimensions are extracted from three data sources: historical transport records (full execution data of similar and route transportation tasks completed by the carrier in the past 1-3 years, which objectively reflects its actual operational level in specific scenarios), transportation equipment information (hardware parameters and status information of vehicles, loading and unloading tools and supporting facilities undertaking this transportation task), and transportation performance data (operational indicators of historical transportation tasks and customer feedback data). After determining the weight coefficients of each indicator through the analytic hierarchy process, the data are weighted and integrated.

[0027] In terms of temperature control capability, the system extracts the accuracy of the temperature control system from the transportation equipment information, namely the maximum deviation between the temperature that the temperature control equipment can stably maintain and the set value, the cooling capacity, whether it is equipped with dual backup refrigeration units, and other hardware parameters. Combined with the temperature compliance rate of goods in the same temperature range in historical transportation records, namely the proportion of the total transportation time with the temperature within the required range, the distribution of the time with temperature exceeding the limit, and the failure rate of refrigeration equipment, the system calculates the temperature control range and temperature fluctuation tolerance that the carrier can stably provide, and also marks whether it has the emergency handling capability for cold chain disruptions.

[0028] In terms of vibration control capability, the system quantifies the maximum vibration control level that the carrier can provide under different modes of transportation and different road conditions based on the vehicle's vibration reduction level, which is determined by the type of suspension system, tire specifications, frame structure, vehicle age, and whether it is equipped with dedicated anti-vibration fixing devices, including air cushion pallets and shock absorber brackets. This is combined with the vibration compliance rate of precision goods in historical transportation records, which is the percentage of road sections where the vibration power spectral density meets the requirements during transportation and the cargo damage rate caused by vibration.

[0029] In terms of delivery timeliness, the system integrates historical on-time rate from transportation fulfillment data, which is the proportion of transportation tasks delivered on time to the total number of tasks, average delay time, and extreme delay probability, which is the proportion of tasks that are delayed beyond the maximum allowable delay time for goods. At the same time, it combines the carrier's capacity scheduling capabilities, including the number of spare vehicles, the processing efficiency of transfer nodes, the average safe driving mileage and violation records of drivers, and the historical travel time data of this transportation route to calculate the shortest delivery time and the maximum delay risk that the carrier can guarantee.

[0030] In terms of loading and unloading operation capabilities, the system extracts the carrier's personnel certification rate, i.e., the proportion of personnel holding professional loading and unloading operation qualification certificates to the total number of loading and unloading personnel, and the configuration rate of special loading and unloading equipment, including the availability of forklifts, cranes, explosion-proof loading and unloading tools, etc. Combined with the loading and unloading accident rate in historical transport records, i.e. the proportion of tasks with cargo damage during loading and unloading, and records of loading and unloading operation violations, the system quantifies these into a loading and unloading operation standard execution capability score corresponding to the requirements of cargo loading and unloading operation standardization, and also indicates whether the carrier has professional loading and unloading qualifications for dangerous goods and oversized and overweight goods.

[0031] In terms of packaging impact resistance, the system quantifies the carrier's ability to protect the packaging during transportation based on the packaging integrity rate in historical transport records (the proportion of goods whose original packaging remains intact after transportation), the mixed loading compliance rate (the proportion of tasks that are strictly loaded according to the type and weight of goods), whether the carrier provides packaging reinforcement services, and the frequency of packaging inspections during transportation.

[0032] Based on this, the system will perform min-max standardization on the quantification results of the five dimensions, convert them into numerical features in the 0-1 range, and then supplement them with binarized or hierarchical features of special capabilities such as dangerous goods transportation qualifications, full-process escort capabilities for valuables, and emergency rescue response time. Finally, the system will splice them together to form a carrying capacity feature vector that is completely consistent with the feature vector dimension of cargo transportation demand. Each component of the vector precisely corresponds to a transportation guarantee capability that can be directly compared with cargo demand.

[0033] S400: Based on the characteristics of cargo transportation demand and carrying capacity, an adaptation deviation analysis is conducted to establish characteristics of transportation mismatch.

[0034] Specifically, the system first establishes a deviation calculation framework for five major transportation assurance dimensions. Each dimension employs a three-level calculation method: basic difference calculation, scenario correction coefficient, and weight allocation. This ensures that the deviation accurately reflects its actual impact on cargo transportation safety. In the temperature control range deviation calculation, the system first extracts the temperature control range required by the cargo, namely the upper and lower temperature limits that must be maintained throughout the cargo transportation process and the temperature range that the carrying capacity can stably provide. The length of the non-overlapping part of the two ranges is calculated as the basic deviation value. Then, a transportation time correction coefficient is introduced, which means that for every 24 hours increase in transportation time, the deviation value is amplified by 1.2 times. This is because the deterioration effect of prolonged temperature exceeding the limit on cargo is cumulative. The temperature sensitivity correction coefficient is inversely proportional to the temperature change tolerance gradient of the cargo. The smaller the temperature change tolerance gradient, the more sensitive the cargo is to temperature deviation, and the larger the correction coefficient. Finally, the temperature control range deviation is obtained. A value of 0 indicates that the carrying temperature range fully covers the demand, a positive value indicates excess carrying capacity, and a negative value indicates insufficient carrying capacity. The larger the absolute value, the more severe the mismatch.

[0035] In calculating the vibration power spectral density deviation, the system compares the required vibration power spectral density of the cargo (i.e., the parameter describing the allowable vibration energy distribution at different frequencies) with the vibration power spectral density curve provided by the transport capacity on a frequency band-by-frequency basis. It focuses on calculating the ±20Hz frequency band near the cargo's natural frequency. When the transport vibration frequency coincides with the cargo's natural frequency, resonance will occur, leading to an exponential increase in the probability of cargo damage. This energy deviation is then multiplied by the vibration sensitivity coefficient corresponding to the cargo's impact fragility value to finally obtain the vibration power spectral density deviation. This dimensional deviation is the core source of mismatch in the transport of fragile goods such as precision instruments and glass products.

[0036] In calculating the delivery timeliness deviation, the system first calculates the difference between the shortest delivery time that the carrier can guarantee and the longest delivery time allowed for the goods as the basic deviation. Then, it introduces an extreme delay correction coefficient, which is proportional to the carrier's extreme delay probability. For every 1% increase in the extreme delay probability, the deviation value is magnified by 1.05 times. It also introduces a timeliness level correction coefficient, which is five times that of a level 3 timeliness goods for special timeliness sensitive goods. Finally, the delivery timeliness deviation is obtained, which directly reflects the risk of loss of goods value due to insufficient carrier timeliness.

[0037] In calculating the deviation of loading and unloading operation standardization, the system calculates the difference between the loading and unloading operation standardization requirements of the goods, namely the quantitative scoring standards for operation actions, tool use, stacking methods, etc. during the loading and unloading process, and the carrier's loading and unloading operation standardization execution capability score. Then, it assigns different weights according to the damage rate of different violations in historical loading and unloading accidents. For example, the weight of the violation of prohibiting inversion is 3 times that of slight throwing. Finally, the deviation of loading and unloading operation standardization is obtained.

[0038] In calculating the packaging impact resistance allowance deviation, the system first calculates the required packaging impact resistance allowance for the goods, which is the difference between the maximum impact protection capability that the packaging can provide and the impact fragility value of the goods, and the difference between the actual packaging protection allowance that the carrier can provide. Then, a mixed loading correction coefficient is introduced. If the carrier has a history of mixing heavy goods with light and fragile goods, the correction coefficient is 1.5. A road condition correction coefficient is also introduced, which is adjusted according to the average terrain ruggedness of the transportation route. The higher the ruggedness, the larger the correction coefficient. Finally, the packaging impact resistance allowance deviation is obtained.

[0039] Based on this, the system will perform z-score standardization on the deviations of the five dimensions, converting them into standardized deviation values ​​with a mean of 0 and a standard deviation of 1. Then, according to the inherent vulnerability parameters of the goods, namely impact brittleness, temperature change tolerance gradient, and value density, the system will automatically assign dynamic weights to each dimension through a pre-trained weight allocation model. For example, for temperature-sensitive goods with a value density exceeding 100,000 yuan / cubic meter, the weight of the temperature control range deviation can reach 0.4, while for ordinary steel goods, the weights of all dimensions are basically equal.

[0040] Finally, the system concatenates the weighted standardized deviation values ​​with the deviation values ​​of special dimensions such as the deviation of explosion protection requirements for dangerous goods and the deviation of full-process monitoring requirements for valuables to form a multi-dimensional transportation mismatch vector. Each component of this vector corresponds precisely to the degree of mismatch in a transportation guarantee dimension, and the magnitude of the vector represents the overall level of transportation mismatch.

[0041] S500: Utilize transportation route data to perform historical risk event correlation analysis on the corresponding transportation segments of the target transportation route, establish the risk amplification trend of transportation mismatch characteristics in the transportation process, and establish transportation risk characteristics.

[0042] Specifically, the system first analyzes the transportation route data, dividing the complete transportation route into transportation segments of 50-100 kilometers in length according to administrative boundaries, road type change points, and abrupt changes in terrain features. This segment serves as the smallest spatial unit for risk assessment, balancing computational efficiency with risk identification accuracy. Each segment is labeled with environmental entity attributes and historical risk tags. The environmental entity attributes are calculated using a digital elevation model to determine terrain ruggedness, i.e., the standard deviation of road elevation (a higher value indicates higher vehicle vibration). The system also includes the probability of seasonal strong winds or rainfall based on meteorological data from the same period over the past five years, i.e., the frequency of extreme weather events, which directly affects equipment reliability and driving stability. Additionally, the system assigns road surface grades according to highway engineering technical standards, which determines the degree of vehicle bumpiness and the probability of accidents. Furthermore, it supplements these attributes with auxiliary attributes such as bridge weight limits, tunnel length, and traffic congestion index. The historical risk tags are extracted from the historical claims database, showing the frequency of cargo damage, cargo discrepancies, and delays in the past three years, the average loss amount, and the main accident types for each segment, forming a standardized baseline probability distribution.

[0043] Based on this, the system constructs a spatiotemporal risk amplification inference network, which is a structural causal model with transportation segments as nodes and segment connection relationships as directed edges. During the model training phase, the system concatenates the environmental entity attributes of historical transportation tasks, historical risk labels, and corresponding multidimensional transportation mismatch vectors. The actual types of risk events and the degree of loss are used as supervision labels. The model parameters are optimized through gradient descent algorithm so that it can accurately learn the amplification rules of different environmental conditions on various types of transportation mismatches. For example, rugged mountain roads will amplify the risk of vibration mismatch by 3-5 times, rainstorms will amplify the risk of temperature mismatch by 2 times, and rainwater can easily cause short circuits in refrigeration equipment.

[0044] After training, the system loads the multi-dimensional transport mismatch vector for this task into the inference network and performs segment-by-segment forward inference sequentially along the transport path. The risk output of the previous segment serves as the initial state input for the next segment, thus simulating the cumulative effect of risk. For example, if vibration in one segment causes packaging to loosen, the probability of damage from vibration in the next segment will significantly increase. Finally, a multi-dimensional risk time-series characteristic curve is generated. The horizontal axis of this curve represents the transport segment sequence, corresponding to transport time, and the vertical axis represents the real-time risk values ​​for five dimensions: temperature, vibration, timeliness, loading and unloading, and packaging. This visually demonstrates the evolution trend of various risks during the transport process. Figure 3 As shown, the risk is amplified by vibration in mountainous sections 4-6, a sudden increase in temperature in rainstorm sections 7-8, and the cumulative effect of risk caused by loose packaging. This verifies the ability of the spatiotemporal risk amplification inference network to characterize the evolution of risk along the path.

[0045] The system then employs a combination of multi-scale one-dimensional convolution and self-attention pooling to extract global risk representations of the route. Multi-scale one-dimensional convolution uses kernels with lengths of 3, 5, and 7 segments to capture risk patterns for short distances, bridge bumps, medium distances, mountainous sections, long distances, and high temperatures throughout the route, respectively. Self-attention pooling automatically assigns dynamic weights to the risk features of each segment, with higher risk values ​​resulting in greater weights. For example, the weight of a high-accident-rate section can be up to 10 times that of an ordinary section. Finally, the system merges these elements to obtain a global risk representation vector that comprehensively reflects the core risks of the entire transportation route.

[0046] Finally, the system parses inherent vulnerability parameters from cargo attribute data, including impact brittleness, temperature tolerance gradient, and value density, which respectively measure the sensitivity of cargo to impact, temperature changes, and economic losses. The system performs an outer product operation on the path global risk representation vector and the inherent vulnerability parameters to generate a high-dimensional transportation risk feature tensor. The outer product operation can achieve cross-combination of each risk dimension and each vulnerability dimension. For example, temperature risk is multiplied by temperature tolerance gradient, and vibration risk is multiplied by impact brittleness, so that the final risk feature can reflect the actual damage probability of a specific cargo under a specific transportation path.

[0047] S600: Based on transportation risk characteristics and historical transportation claims data, perform similar scenario matching, use the similar scenario matching results to perform insurance coverage matching, and output insurance matching results.

[0048] Specifically, the high-dimensional transportation risk feature tensor is first flattened into a one-dimensional risk feature vector in a fixed-dimensional order. This vector contains 128 dimensions of risk information, including inherent cargo vulnerability, global route risk, and mismatch risk across various transportation guarantee dimensions. It serves as a standardized carrier for measuring the risk similarity of different transportation tasks. Simultaneously, one-dimensional historical risk feature vectors corresponding to all closed transportation tasks from the past five years are extracted from the historical transportation claims database to construct a risk feature index library containing millions of samples. Pre-trained dynamic weight coefficients are loaded, and sample pairs are constructed by matching real insurance types and claims results in historical insurance matching records. The optimization objective is to maximize the correlation between matching similarity and insurance matching accuracy. Adjustable weight values ​​are obtained through iterative learning using a gradient descent algorithm. For example, for cold chain vaccine transportation tasks, the weight of the temperature risk dimension can reach 0.35, while for ordinary steel transportation tasks, the weight of this dimension is only 0.05. A corresponding dynamic weight is assigned to each component of the one-dimensional risk feature vector.

[0049] The system then calculates the weighted cosine similarity between the weighted target risk feature vector and the historical risk feature vector, extracting K historical transportation tasks with a similarity higher than a preset threshold, typically 0.85, as the initial set of similar scenarios. K-means clustering analysis is then performed on the claims results of these K historical tasks in the initial set to identify high-frequency loss types, including corrosion losses due to excessive temperature, damage losses due to vibration and impact, and time-sensitive losses due to traffic congestion. A risk-dominant label is established based on the distribution ratio of each loss type across the K tasks. When a certain loss type accounts for more than 60%, it is determined as the dominant loss pattern for this transportation task. This risk-dominant label is then used to filter the initial set of similar scenarios, removing historical tasks whose loss type deviates from the dominant label by more than 30%. Finally, a precise matching result set consisting of 20-30 highly similar scenarios is formed.

[0050] Based on this, the system executes an insurance coverage matching process. First, it statistically analyzes the historical compensation amounts and loss coverage of different insurance types in the precise matching result set, and calculates the insurance type loss compensation capability parameter, the ratio of the cumulative compensation amount to the cumulative actual loss amount of the insurance type in similar scenarios (reflecting the actual compensation ratio of the insurance type for similar losses; the higher the ratio, the stronger the compensation capability), and the insurance type risk coverage capability parameter, the ratio of the number of loss types covered by the insurance type to the total number of potential loss types in the current task in similar scenarios (reflecting the breadth of coverage of various risks). Then, the system assigns dynamic weights to the two parameters according to the risk-dominant label of the current transportation task. For example, for temperature-dominant cargo damage scenarios, the weight of the loss compensation capability parameter is set to 0.7, and the weight of the risk coverage capability parameter is set to 0.3. For multi-risk mixed scenarios, the weight of both parameters is set to 0.5. Through weighted fusion calculation, the insurance type coverage contribution value of each insurance type is obtained, which is the core indicator for comprehensively measuring the risk protection value of the insurance type for the current transportation task.

[0051] Finally, the system sorts all insurance types from highest to lowest coverage contribution value, selects the top 3 main insurances and 2 supplementary insurances, and calculates the recommended insured amount and estimated premium based on the total value of goods, risk level and historical average loss ratio of this task. At the same time, it marks the coverage scope, exclusions and claims precautions of each insurance type, and finally forms a standardized insurance matching result that includes three parts: the optimal insurance combination plan, alternative plans and risk warnings.

[0052] Furthermore, in the method provided in the application embodiment, an adaptation deviation analysis is performed based on the characteristics of cargo transportation demand and the characteristics of carrying capacity to establish transportation mismatch degree characteristics, including: establishing a transportation assurance dimension, wherein the transportation assurance dimension includes temperature control range deviation, vibration power spectral density deviation, delivery timeliness deviation, loading and unloading operation standardization deviation, and packaging impact resistance margin deviation; under the transportation assurance dimension, the characteristics of cargo transportation demand and the characteristics of carrying capacity are subjected to dimension-by-dimensional difference calculation to establish a multi-dimensional transportation mismatch vector, and the multi-dimensional transportation mismatch vector is output as the transportation mismatch degree characteristics.

[0053] Specifically, the system first establishes five transportation guarantee dimensions that correspond one-to-one with demand analysis and capacity analysis, ensuring that the deviation calculation for each dimension has a clear supply-demand correspondence and that there is no misalignment between dimensions. Then, differentiated difference calculations are performed under each dimension: In the calculation of temperature control range deviation, the temperature control range deviation refers to the temperature control range required by the goods, that is, the upper and lower limits of the temperature that must be maintained throughout the transportation process, such as 2-8℃ for cold chain vaccines and below -18℃ for frozen seafood. The system calculates the length of the non-overlapping part of the temperature range that the carrier can stably provide. The system first calculates the intersection length of the two ranges, and then subtracts the intersection length from the total length of the demand range to obtain the basic deviation value. If the carrier range completely includes the demand range, the deviation value is 0 or even negative, indicating excess capacity. If there is no intersection, the deviation value is equal to the total length of the demand range, indicating a complete mismatch. At the same time, the deviation value is cumulatively corrected according to the transportation time. For every 24 hours increase in transportation time, the deviation value is amplified by 1.2 times to reflect the cumulative deterioration effect of temperature exceeding the limit on the goods.

[0054] In the calculation of vibration power spectral density deviation, the vibration power spectral density deviation is the sum of vibration energy differences after weighting each frequency band. The system compares the vibration power spectral density requirement of the cargo, that is, the parameter describing the allowable vibration energy distribution at different frequencies, reflecting the sensitivity of the cargo to vibrations at different frequencies, with the vibration power spectral density curve provided by the carrying capacity at a precision of 1Hz. It gives more than 5 times the weight to the resonant sensitive frequency band within ±20Hz of the cargo's natural frequency. Finally, the deviation value of this dimension is obtained by weighted summation, avoiding the defect that the traditional single vibration acceleration index cannot reflect frequency sensitivity.

[0055] In the calculation of delivery time deviation, the delivery time deviation is the 95th percentile delivery time of the carrier, that is, the longest delivery time that 95% of historical transportation tasks can achieve. It is specifically used to cover the risk of extreme delays with low probability but high loss. It is the difference between the delivery time deviation and the maximum allowable delay time of the goods. That is, the critical time when the value of the goods will be completely or partially lost after the delivery time exceeds the value, such as 2 hours for emergency medicines and 12 hours for fresh fruits. The 95th percentile time is used instead of the average time to avoid the average data from masking the significant losses caused by extreme events.

[0056] In calculating the deviation of loading and unloading operation standardization, the deviation is the difference between the minimum required standardization score for cargo loading and unloading and the actual performance score of the carrier. The standardization of loading and unloading operations is a quantitative scoring system comprising 12 sub-items, including compliance of operational actions, correct use of tools, reasonable stacking methods, and personnel qualification compliance rate, with a maximum score of 100. The system assigns differentiated weights based on the historical damage rate of different violations. For example, the weight for the violation of prohibiting inversion is three times that of minor stacking violations, and the weight for the violation of throwing goods is twice that of manual handling exceeding weight limits. Regarding packaging impact resistance allowance... In the deviation calculation, the packaging impact resistance margin deviation is the minimum packaging impact resistance margin required for the goods. It is the difference between the maximum impact protection capability that the packaging can provide and the impact fragility value of the goods. The larger the margin, the higher the impact resistance safety of the goods. It is the difference between the packaging impact resistance margin and the actual packaging protection margin that the carrier can provide. The carrier's packaging protection margin includes not only the protection capability of the goods' own packaging, but also the contribution of additional reinforcement, padding, isolation and other transportation protection measures provided by the carrier. At the same time, the deviation value will be corrected according to the carrier's historical mixed loading violation records. If there is a history of mixing heavy goods with light and fragile goods, the deviation value will be magnified by 1.5 times.

[0057] After completing the difference calculations for the five dimensions, the system performs z-score standardization on all deviation values, converting them into standardized values ​​with a mean of 0 and a standard deviation of 1. This eliminates the incomparability caused by differences in the dimensions. Then, the five standardized deviation values ​​are concatenated in a fixed order of temperature, vibration, timeliness, loading and unloading, and packaging to form a multi-dimensional transportation mismatch vector. Each component of this vector precisely corresponds to the degree of mismatch in a transportation support dimension. Positive values ​​indicate insufficient carrying capacity, while negative values ​​indicate excessive carrying capacity. The larger the absolute value, the more severe the mismatch. The magnitude of the vector represents the overall comprehensive mismatch level of this transportation task.

[0058] Furthermore, the method provided in the application embodiment establishes transportation risk characteristics, including: parsing the transportation route data, extracting N consecutive transportation segments traversed by the target transportation route, and labeling each transportation segment with environmental entity attributes and historical risk tags. The environmental entity attributes include the segment's terrain ruggedness, probability of seasonal strong winds or rainfall, and road surface grade. The historical risk tags include the baseline probability distribution of cargo damage, cargo discrepancies, and delay events in the segment. A spatiotemporal risk amplification inference network is established, where the spatiotemporal risk amplification inference network uses transportation segments as nodes and the connection relationships between transportation segments as edges, including mismatch condition input... The system employs a structural causal model for inputting the interface; it concatenates the environmental entity attributes with historical risk labels and inputs this concatenation into the spatiotemporal risk amplification inference network for supervised training of network parameters; it loads a multidimensional transportation mismatch vector into the trained spatiotemporal risk amplification inference network and performs forward inference segment by segment along the path defined by the segment attribute sequence to establish a multidimensional risk temporal feature curve reflecting the evolution of risk along the path; it performs multi-scale one-dimensional convolution and self-attention pooling operations on the multidimensional risk temporal feature curve to extract the path global risk representation vector; and it establishes transportation risk features based on the path global risk representation vector.

[0059] Specifically, the system first performs multi-dimensional analysis of the transportation route data, dividing the complete transportation route into transportation segments of 50-100 kilometers in length according to administrative boundaries, road type change points, and abrupt changes in terrain features. This is the smallest spatial unit for risk assessment. This granularity balances computational efficiency and risk identification accuracy; too coarse a granularity will mask local high-risk points, while too fine a granularity will introduce excessive noise and increase computational burden. Each transportation segment is labeled with environmental entity attributes and historical risk tags. The environmental entity attributes are calculated using a 10-meter resolution digital elevation model to obtain terrain ruggedness, i.e., the standard deviation of road segment elevation. The higher the value, the higher the vibration amplitude and bump frequency of vehicle movement. The data is also based on hourly meteorological data from the same period over the past five years. The probability of seasonal strong winds or rainfall, i.e. the frequency of extreme weather events, directly affects the reliability and stability of transportation equipment. The road surface grades, classified according to highway engineering technical standards (five grades from expressways to gravel roads), determine the basic bumpiness of vehicle travel and the probability of accidents. It is supplemented by auxiliary attributes such as bridge weight limits, tunnel length, and daily traffic congestion index. The historical risk label extracts data on all settled cargo damage, cargo loss, and delay events in the past three years from the insurance company's historical claims database. It statistically obtains the frequency of occurrence, average loss amount, and main causes of accidents for each type of event, forming a standardized baseline probability distribution that reflects the inherent risk level of the section under the condition of no special mismatch.

[0060] Based on this, the system constructs a spatiotemporal risk amplification inference network, which is a structural causal model with transportation segments as nodes and the physical connections between segments as directed edges. This model can clearly reveal the causal transmission relationship between environmental attributes, mismatch degree, and risk events. The network is designed with a mismatch condition input interface to receive the multi-dimensional transportation mismatch vector generated in the preceding sequence. During the model training phase, the system concatenates the environmental entity attributes, historical risk labels, and multi-dimensional transportation mismatch vectors corresponding to historical transportation tasks. The actual risk event type, occurrence time, and loss degree are used as supervision labels. The model parameters are iteratively optimized using the Adam gradient descent algorithm, enabling it to accurately learn the differentiated amplification laws of various transportation mismatches under different environmental conditions. For example, rugged mountain roads will amplify the risk of vibration mismatch by 3-5 times, continuous heavy rain will amplify the risk of temperature mismatch by 2 times, rainwater can easily cause short circuits in refrigerated truck refrigeration equipment, and congested highway sections will amplify the risk of timeliness mismatch by 4 times.

[0061] After training, the system loads the multidimensional transportation mismatch vector of this task into the inference network and performs segment-by-segment forward inference sequentially along the transportation path. The risk output of the previous segment is used as the initial state input of the next segment to simulate the cumulative effect of risk. For example, if severe vibration in the previous segment causes the packaging of the goods to loosen, the probability of damage caused by vibration of the same intensity in the next segment will increase by 2-3 times. Finally, a multidimensional risk time series feature curve is generated. The horizontal axis of the curve is the order of transportation segments, which corresponds to the transportation time, and the vertical axis is the real-time risk value of five dimensions: temperature, vibration, timeliness, loading and unloading, and packaging. It can intuitively show the evolution trend and peak position of various risks as they change during the transportation process.

[0062] The system then employs a combination of multi-scale one-dimensional convolution and self-attention pooling to extract global risk representations of the route. The multi-scale one-dimensional convolution uses kernels with segment lengths of 3, 5, and 7, respectively. The 3-segment kernel captures short-distance sudden risks, such as the turbulence risk of a single bridge; the 5-segment kernel captures medium-distance continuous risks, such as the vibration risk of continuous mountain road sections; and the 7-segment kernel captures long-distance systemic risks, such as the temperature risk of a high-temperature environment throughout the route. The self-attention pooling automatically assigns dynamic weights to the risk features of each segment, with higher risk values ​​resulting in greater weights. Finally, the system merges these weights to obtain a global risk representation vector that comprehensively reflects the core risks of the entire transportation route.

[0063] Finally, the system extracts inherent vulnerability parameters from cargo attribute data, including impact brittleness, temperature tolerance gradient, and value density, which respectively measure the sensitivity of cargo to impact, temperature changes, and economic losses. The system performs an outer product operation on the path global risk representation vector and the inherent vulnerability parameters to generate a high-dimensional transportation risk feature tensor. The outer product operation can achieve cross-combination of each risk dimension and each vulnerability dimension. For example, temperature risk is multiplied by temperature tolerance gradient, and vibration risk is multiplied by impact brittleness, so that the final risk feature can accurately reflect the actual damage probability of a specific cargo under a specific transportation path.

[0064] Furthermore, in the method provided in the application embodiment, establishing transportation risk features based on the path global risk representation vector includes: parsing the inherent vulnerability parameters of the target cargo based on cargo attribute data, wherein the inherent vulnerability parameters include impact brittleness value, temperature change tolerance gradient, and value density; performing an outer product operation on the path global risk representation vector and the inherent vulnerability parameters to generate a high-dimensional transportation risk feature tensor; and outputting the high-dimensional transportation risk feature tensor as the transportation risk feature.

[0065] Specifically, the system first analyzes the inherent vulnerability parameters of the target cargo based on standardized cargo attribute data. These parameters are a set of quantitative indicators describing the sensitivity of the cargo's physical, chemical, and economic properties to external risks, comprising three independent and complementary dimensions: The first is impact fragility value, which refers to the maximum impact acceleration the cargo can withstand, measured in g. Exceeding this value will cause irreversible structural damage or performance failure. The system uses OCR to identify relevant parameters in the cargo manual and product inspection reports, or matches corresponding values ​​from a pre-trained cargo category-fragility value mapping database. For example, the impact fragility value of precision optical instruments is typically 10-30g, that of ordinary electronic products is 50-100g, while that of bulk commodities such as steel can reach over 500g. The lower the impact fragility value, the more sensitive the cargo is to vibration and impact risks during transportation. The second dimension is temperature resistance gradient. Temperature tolerance refers to the rate at which goods deteriorate under unit temperature change, expressed as % / ℃. The system calculates this rate based on the material characteristics, shelf-life requirements, and industry standards of the goods. For example, the temperature tolerance gradient for cold chain vaccines is approximately 5% / ℃, meaning that for every 1℃ deviation from the required range, the vaccine potency decreases by 5% per hour. In contrast, the temperature tolerance gradient for ordinary room-temperature goods can be as low as 0.1% / ℃ or less. The greater the temperature tolerance gradient, the more sensitive the goods are to temperature fluctuations. Thirdly, value density refers to the economic value of goods per unit volume or weight, expressed as yuan / cubic meter or yuan / kilogram. This value density is automatically calculated by the system based on the total value, volume, and weight of the goods. For example, the value density of chips can reach tens of millions of yuan per cubic meter, while the value density of coal is only a few hundred yuan per cubic meter. Value density directly determines the scale of economic loss caused by the same degree of cargo damage.

[0066] After analyzing the three inherent vulnerability parameters, the path global risk representation vector and the inherent vulnerability parameter vector are multiplied by an outer product operation, also known as a tensor product operation. This operation achieves pairwise cross-combinations between all dimensions of the two vectors, capturing all interactions between risk and vulnerability. Assuming the path global risk representation vector is m-dimensional, it typically includes five core risk dimensions (temperature, vibration, timeliness, loading and unloading, and packaging) and multiple auxiliary risk dimensions. The inherent vulnerability parameters are 3-dimensional vectors. The outer product operation will generate an m×3 two-dimensional high-dimensional transportation risk feature tensor. Each element in the tensor corresponds to the cross-influence value of a specific risk dimension and a specific vulnerability parameter. For example, the product of the temperature risk dimension and the temperature tolerance gradient represents the actual temperature-induced damage risk of the goods on this transportation route; the product of the vibration risk dimension and the impact brittleness value represents the actual vibration-induced damage risk; and the product of the timeliness risk dimension and the value density represents the economic loss risk caused by actual timeliness delays.

[0067] The outer product operation method can reflect the differentiated response of different goods to the same environmental risk. For example, the same temperature fluctuation risk will have a very high actual risk to cold chain vaccines with a large temperature change tolerance gradient, while it will have almost no impact on ordinary goods with a small temperature change tolerance gradient; the same vibration and shock risk will cause serious damage to precision instruments with low impact brittleness value, while it can be completely ignored for steel with high impact brittleness value.

[0068] Finally, this high-dimensional transportation risk feature tensor, which contains all the information about risk-vulnerability interactions, is output as the transportation risk feature. This feature represents the actual risk level of a specific cargo in a specific route environment during this transportation mission.

[0069] Furthermore, in the method provided in the application embodiment, similar scenario matching based on transportation risk features and historical transportation claims data includes: flattening the transportation risk features into a one-dimensional risk feature vector, and simultaneously extracting one-dimensional historical risk feature vectors corresponding to each historical transportation task from the historical transportation claims data; assigning a learnable dynamic weight coefficient to each component of the one-dimensional risk feature vector; calculating a weighted cosine similarity based on the weighted one-dimensional risk feature vector and the weighted one-dimensional historical risk feature vector respectively, and extracting K historical transportation tasks with similarity higher than a preset threshold as similar scenario matching results for output.

[0070] Specifically, the process begins with feature flattening. The m×3 high-dimensional transportation risk feature tensor, which is a two-dimensional tensor generated by the cross-product of the path global risk representation vector and the inherent vulnerability parameter, contains the cross-influence information of all risk and vulnerability dimensions. It is then expanded row by row in a fixed order: temperature risk × temperature change tolerance gradient, temperature risk × impact brittleness value, temperature risk × value density, vibration risk × temperature change tolerance gradient, etc., and converted into a 3m-dimensional one-dimensional risk feature vector. The flattening process strictly maintains the consistency of the dimensional order, ensuring that all historical and current task feature vectors have the same dimensional structure and physical meaning. The synchronization system then extracts high-dimensional risk feature tensors corresponding to all closed transportation tasks from the distributed historical claims database in batches. These tensors are flattened into one-dimensional historical risk feature vectors according to the same rules, and a Faiss high-efficiency vector index library containing millions of samples is constructed, supporting millisecond-level similarity retrieval.

[0071] Next, the system assigns learnable dynamic weight coefficients to each component of the one-dimensional risk feature vector. This is a set of weight values ​​automatically optimized through machine learning, which can automatically adjust the importance of each dimension according to the risk characteristics of different transportation tasks. The learning process is based on historical insurance matching records: the system extracts real insurance matching schemes and corresponding claims results from history to construct sample pairs. With the optimization objective of maximizing matching similarity and being positively correlated with insurance matching accuracy, the system iteratively adjusts the weights of each dimension through the Adam gradient descent algorithm, so that the risk dimensions that have a greater impact on insurance decisions receive higher weights. For example, for cold chain vaccine transportation tasks, the weight of the temperature risk × temperature change tolerance gradient component can reach 0.35, while for ordinary steel transportation tasks, the weight of this component is only 0.05, and the weight of the vibration risk × impact brittleness component will be increased accordingly.

[0072] After completing the weight allocation, the system calculates the weighted cosine similarity based on the weighted target risk feature vector and the weighted historical risk feature vector. During the calculation, the corresponding components of the two vectors are first multiplied by their respective dynamic weight coefficients, and then the ratio of the dot product of the weighted vectors to the product of their moduli is calculated. The similarity value ranges from [-1, 1]. The closer the value is to 1, the more similar the risk patterns of the two transportation tasks are.

[0073] Finally, the system retrieves all historical transportation tasks with similarity higher than a preset threshold from the vector index library. The threshold is usually set to 0.85, which is determined through cross-validation to balance the relevance of similar scenarios with the sample size. If the number of tasks that meet the criteria exceeds the preset K value (K=50 by default), the top K tasks are extracted as similar scenario matching results after being sorted from high to low similarity. If the number of tasks that meet the criteria is less than K, all tasks with similarity higher than the threshold are extracted to avoid introducing irrelevant data with large differences in risk patterns in order to make up the sample size.

[0074] Furthermore, in the method provided in the application embodiment, the configuration of dynamic weight coefficients includes: obtaining real insurance type selection and claim results from historical insurance matching records to construct sample pairs; taking the correlation between maximizing matching similarity and insurance matching accuracy as the optimization objective, learning the weights of risk features in each dimension through gradient descent method, and establishing dynamic weight coefficients.

[0075] Specifically, the system first extracts all historical insurance matching records that have been closed within the past 5 years from the distributed insurance business database. This includes complete business files containing real insurance information, claims results, and corresponding original transportation data. It covers the entire process information, including the actual insurance combination purchased by the cargo owner, the insured amount, the premium, the time and place of the accident, the type of loss, the amount of loss, the final compensation amount, and the claims processing time. At the same time, it removes insurance fraud cases, extreme claims cases caused by force majeure, and invalid records with more than 30% missing data to ensure the authenticity and reliability of the training data.

[0076] Based on this, the system constructs training sample pairs. Each sample pair consists of two parts: first, a one-dimensional risk feature vector corresponding to the historical transportation task, which is completely consistent with the feature vector dimension of the current task; and second, an insurance matching quality label calculated based on the actual claims results. If the actual payout rate of a historical task, i.e., the payout amount / actual loss amount, is in a reasonable range of 80%-120%, and all types of losses are covered by the insured, it is marked as an accurate match with a label value of 1. If the payout rate is lower than 80%, it is marked as underinsured with a label value of 0. If it is higher than 120%, it is marked as overinsured with a label value of 0. Finally, a training dataset containing tens of thousands of high-quality samples is formed.

[0077] Next, the objective function is to maximize the positive correlation between weighted cosine similarity and insurance matching accuracy. Specifically, the system calculates the weighted cosine similarity between any two samples in the training set, and at the same time calculates the consistency of the insurance matching quality labels of the two samples. That is, the more similar the risk patterns of two transportation tasks are, the more similar their optimal insurance schemes are. Here, the Pearson correlation coefficient is used to quantify this correlation, and the goal is to maximize the correlation coefficient.

[0078] The system then uses the Adam gradient descent method to iteratively learn the weights of risk features in each dimension. First, the learnable dynamic weight coefficients are initialized as uniformly distributed random vectors of the same dimension as the risk feature vectors, and normalized so that the sum of all weights is 1. Then, it enters the iterative training process. In each iteration, the system first calculates the weighted cosine similarity of all sample pairs based on the current weights, then calculates the Pearson correlation coefficient between the similarity and the consistency of the insurance matching quality label as the current objective function value. Next, it calculates the gradient of the objective function with respect to each weight parameter using the backpropagation algorithm, updating the weight parameters along the direction of gradient ascent to maximize relevance. The training process continues until the change in the objective function value is less than 1e-6, i.e., convergence or reaching the preset maximum number of iterations. Finally, an optimal set of dynamic weight coefficients is obtained. This set of weights can automatically distinguish the importance of different risk dimensions to insurance decisions. For example, for cold chain vaccine transportation tasks, the weight of the "temperature risk × temperature change tolerance gradient dimension" can reach 0.35, while for ordinary steel transportation tasks, the weight of this dimension is only 0.05, and the weight of the "vibration risk × impact brittleness value dimension" will correspondingly increase to over 0.25.

[0079] Furthermore, in the method provided in the application embodiment, extracting K historical transportation tasks with similarity higher than a preset threshold as similar scenario matching results output includes: performing cluster analysis on the historical claims results corresponding to the K historical transportation tasks to identify high-frequency loss types in the claims results; establishing risk-dominant labels based on the distribution ratio of high-frequency loss types in the K historical transportation tasks; using the risk-dominant labels to perform similar scenario filtering on the K historical transportation tasks, eliminating historical transportation tasks whose loss types deviate from the risk-dominant labels, and establishing similar scenario matching results.

[0080] Specifically, the system first performs structured feature processing on the historical claims results corresponding to K historical transportation tasks, converting the loss information in each claims report into a standardized loss type vector: first, it extracts information such as accident cause, damaged part, and loss nature from the claims data, and maps it to 8 predefined basic loss types, such as temperature corrosion, vibration and impact damage, time delay depreciation, loading and unloading operation damage, packaging damage and leakage, traffic accident damage, and theft and loss. The value of each dimension is the proportion of the loss amount of this type to the total loss amount of this accident, forming an 8-dimensional loss type feature vector. For example, in the claims results of a certain cold chain transportation task, the loss due to temperature corrosion accounts for 92% and the loss due to packaging damage accounts for 8%, and its loss type vector is [0.92,0,0,0,0.08,0,0,0]. Based on this, the system uses the K-means clustering algorithm to perform cluster analysis on K loss type vectors. The optimal number of clusters is determined by the elbow rule, which is usually 2-3. The initial set is divided into several clusters with similar loss patterns, and each cluster corresponds to a typical loss combination pattern.

[0081] The system then counts the number of samples and the cumulative loss amount for each cluster, calculates the frequency and cumulative loss amount of each basic loss type in all K tasks, and identifies high-frequency loss types, i.e. loss types with a frequency exceeding 60% or a cumulative loss amount exceeding 70%. For example, in 50 initial similar tasks, 42 tasks experienced vibration and impact damage, with the cumulative loss amount accounting for 83% of the total loss. Therefore, vibration and impact damage is the high-frequency loss type for this transportation task.

[0082] Next, the system establishes risk-dominant labels based on the distribution characteristics of high-frequency loss types. If there is only one high-frequency loss type that meets the conditions, a single risk-dominant label is established, including temperature-dominant cargo loss, vibration-dominant cargo loss, etc. If there are two high-frequency loss types and their cumulative loss amount accounts for more than 85%, a mixed risk-dominant label is established, including vibration and time-related cargo loss, etc. If there is no obvious high-frequency loss type, a multi-risk mixed label is established, and the weight of each major loss type is recorded.

[0083] Finally, the system uses risk-dominant labels to filter the initial K historical transportation tasks. For single-type dominant labels, it retains all historical tasks whose claims results contain the dominant loss type and whose loss accounts for more than 30%, and removes pseudo-similar tasks that do not contain the type at all or whose loss accounts for less than 30%. For mixed-type dominant labels, it retains tasks that contain two dominant loss types or at least one type and whose loss accounts for more than 40%. For multi-risk mixed labels, it removes only extreme abnormal tasks whose loss type has no intersection with any of the major risk types. After filtering, the system checks the number of remaining valid samples. If the sample size is less than 15 to ensure the statistical significance of subsequent insurance type statistical analysis, it appropriately lowers the preset similarity threshold by 0.02 each time and re-searches and supplements similar tasks until the sample size meets the requirements, ultimately forming a precise similar scenario matching result set consisting of 15-30 highly homogeneous historical tasks.

[0084] Furthermore, in the method provided in the application embodiment, insurance coverage matching is performed using similar scenario matching results, and insurance matching results are output. This includes: statistically analyzing the historical compensation amounts and loss coverage ranges corresponding to different insurance types in the similar scenario matching results; establishing insurance type loss compensation capability parameters based on historical compensation amounts, and establishing insurance type risk coverage capability parameters based on loss coverage ranges; performing weighted fusion calculations on the loss compensation capability parameters and risk coverage capability parameters of each insurance type based on the dominant loss mode corresponding to the target cargo transportation task, and establishing insurance type coverage contribution values; and outputting the insurance matching results corresponding to the target transportation task according to the insurance type coverage contribution value ranking results.

[0085] Specifically, the system first performs a full-scale structured statistical analysis of all closed claims records in the similar scenario results set, extracting the historical payout amount and loss coverage for each insurance product currently on sale. The historical payout amount refers to the total amount of compensation actually paid by this insurance product for all accidents that meet the claim conditions in similar scenarios, including all payouts such as direct cargo damage compensation, reasonable salvage costs, and third-party liability compensation. During the statistics, cases of insurance fraud, zero-payout cases caused by absolute deductibles, and extreme payout cases caused by force majeure will be excluded to avoid parameter distortion. The loss coverage refers to the set of loss types that have actually been successfully compensated for by this insurance product in similar scenarios, rather than the theoretical coverage scope in the insurance terms. This is because there are often inconsistencies between the terms and practices in actual claims. For example, although some basic insurance terms specify coverage for accidental events, in practice, damage caused by improper loading and unloading operations is often not compensated.

[0086] Based on this, the system establishes two core insurance assessment parameters. The first is the insurance type's loss compensation capability parameter, calculated by dividing the cumulative compensation amount of this type of insurance in similar scenarios by the cumulative actual loss amount of the corresponding accident. This parameter directly reflects the actual compensation ratio of the insurance type for similar losses. For example, in similar scenarios of cold chain transportation, the loss compensation capability parameter for refrigerated cargo transportation insurance is typically 0.85-0.95, while the compensation capability parameter for temperature-induced spoilage loss under general cargo transportation basic insurance is only 0.1-0.2, truly reflecting the differences in compensation capability of different insurance types for specific losses. The second is the insurance type... The risk coverage parameter is calculated as the ratio of the number of loss types covered by this insurance in similar scenarios to the total number of potential loss types in this transportation mission. It is also weighted and adjusted according to the risk weight of each loss type. For example, for a vibration-dominated cargo damage scenario, the weight of vibration impact loss type is set to 0.6, and the weight of other loss types is 0.1. If an insurance covers both vibration impact and packaging damage loss types, its risk coverage parameter is (0.6+0.1) / 1=0.7, rather than a simple 2 / 5, which more accurately reflects the insurance's coverage of core risks.

[0087] Next, the system assigns dynamic weights to two parameters based on the dominant loss pattern of this transportation mission, namely the risk-dominant label established in the previous steps. For a single dominant loss pattern, including temperature-dominant cargo damage, the weight of the loss compensation capability parameter is set to 0.7, and the weight of the risk coverage capability parameter is set to 0.3, indicating that the requirement is full compensation for the dominant loss. For a mixed dominant loss pattern, including vibration and time-related cargo damage, the weight of both parameters is set to 0.5, balancing the intensity of compensation and the breadth of coverage. For a multi-risk mixed pattern, the weight of the risk coverage capability parameter is appropriately increased to 0.6 to ensure that all major risks are effectively covered. The system then calculates the coverage contribution value of each type of insurance through weighted fusion, which is an indicator that comprehensively measures the risk protection value of each type of insurance for this transportation mission. A higher contribution value indicates a higher cost-effectiveness of the insurance.

[0088] Finally, the system sorts all available insurance products from highest to lowest coverage contribution value, selects the top 3 main insurance products and 2 supplementary insurance products, and calculates the recommended insured amount based on the total value of the goods in this task, the historical average loss ratio for similar scenarios, and the maximum single loss amount. This recommended amount is typically 1.1-1.5 times the total value of the goods to cover rescue costs and reasonable indirect losses, along with the estimated premium. The system also marks the coverage scope, exclusions, deductible percentage, and claims process for each insurance product. Ultimately, a standardized insurance matching result is generated, which includes the optimal insurance combination plan, cost-effective alternative plans, and risk warnings. The result also lists the historical claims data for each insurance product in similar scenarios as a reference for decision-making.

[0089] In summary, the machine learning-based cargo transportation insurance matching method provided in this application has the following technical effects: By structurally analyzing and feature-engineering multi-source heterogeneous transportation data, a dimensional adaptation deviation quantification system for cargo transportation demand and carrying capacity is established. A spatiotemporal risk amplification inference network based on a structural causal model is constructed to accurately depict the evolution of risk along the transportation path. Combined with weighted cosine similarity, accurate retrieval of similar claims scenarios and quantitative assessment of the actual coverage of insurance types are achieved. Finally, a personalized insurance combination scheme that highly matches the actual transportation risks is output, effectively reducing the cost of cargo transportation insurance and improving the protection efficiency and resource utilization of cargo transportation insurance.

[0090] Example 2, based on the same inventive concept as the machine learning-based cargo transportation insurance matching method in the foregoing examples, such as... Figure 2 As shown in the embodiment of this application, a cargo transportation insurance matching system based on machine learning is provided. The system includes: The module 11 reads transportation data for the target cargo transportation task, including cargo attribute data, transportation route data, carrier data, and historical transportation claims data. The module 12 analyzes the cargo attribute data to determine transportation demand, establishing cargo transportation demand characteristics that characterize the target cargo's requirements for transportation conditions during transportation. The module 13 executes the carrier data to analyze transportation capacity, using historical transportation records, transportation equipment information, and transportation performance data to establish carrier capacity characteristics that characterize the carrier's actual transportation capacity. The module 14 performs adaptation deviation analysis based on cargo transportation demand characteristics and carrier capacity characteristics to establish transportation mismatch characteristics. The module 15 performs historical risk event correlation analysis on the transportation route corresponding to the target transportation segment using transportation route data, establishing the risk amplification trend of transportation mismatch characteristics during transportation and establishing transportation risk characteristics. The module 16 matches transportation risk characteristics with historical transportation claims data to perform similar scenario matching, uses the similar scenario matching results to perform insurance coverage matching, and outputs insurance matching results.

[0091] Furthermore, the analysis module 14 is also used to perform the following steps to establish a transportation assurance dimension, which includes temperature control range deviation, vibration power spectral density deviation, delivery timeliness deviation, loading and unloading operation standardization deviation, and packaging impact resistance margin deviation; under the transportation assurance dimension, the cargo transportation demand characteristics and carrying capacity characteristics are subjected to dimension-by-dimensional difference calculation to establish a multi-dimensional transportation mismatch vector, and the multi-dimensional transportation mismatch vector is output as a transportation mismatch degree feature.

[0092] Furthermore, the risk module 15 is also used to perform the following steps: parse the transportation route data, extract N consecutive transportation segments along the target transportation route, and label each transportation segment with environmental entity attributes and historical risk tags. The environmental entity attributes include the segment's terrain ruggedness, probability of seasonal strong winds or rainfall, and road surface grade. The historical risk tags include the baseline probability distribution of cargo damage, cargo discrepancies, and delay events in the segment. A spatiotemporal risk amplification inference network is established, with transportation segments as nodes and the connections between transportation segments as edges, including a mismatch condition input interface. The structural causal model is constructed; the environmental entity attributes and historical risk labels are concatenated and then input into the spatiotemporal risk amplification inference network for supervised training of network parameters; the multidimensional transportation mismatch vector is loaded into the trained spatiotemporal risk amplification inference network, and forward inference is performed segment by segment along the path direction defined by the segment attribute sequence to establish a multidimensional risk time series feature curve reflecting the evolution of risk along the path; multi-scale one-dimensional convolution and self-attention pooling operations are performed on the multidimensional risk time series feature curve to extract the path global risk representation vector; transportation risk features are established based on the path global risk representation vector.

[0093] Furthermore, the risk module 15 is also used to perform the following steps: parse the inherent vulnerability parameters of the target cargo based on cargo attribute data, the inherent vulnerability parameters including impact brittleness value, temperature change tolerance gradient, and value density; perform an outer product operation on the path global risk representation vector and the inherent vulnerability parameters to generate a high-dimensional transportation risk feature tensor, and output the high-dimensional transportation risk feature tensor as a transportation risk feature.

[0094] Furthermore, the matching module 16 is also used to perform the following steps: flattening the transportation risk features into a one-dimensional risk feature vector, and simultaneously extracting the one-dimensional historical risk feature vector corresponding to each historical transportation task from the historical transportation claims data; assigning a learnable dynamic weight coefficient to each component of the one-dimensional risk feature vector; calculating the weighted cosine similarity based on the weighted one-dimensional risk feature vector and the weighted one-dimensional historical risk feature vector respectively, and extracting K historical transportation tasks with similarity higher than a preset threshold as similar scenario matching results for output.

[0095] Furthermore, the matching module 16 is also used to perform the following steps: obtain the real insurance type selection and claim results in the historical insurance matching records to construct sample pairs; take the correlation between maximizing the matching similarity and the insurance matching accuracy as the optimization objective, learn the weights of risk features in each dimension through the gradient descent method, and establish dynamic weight coefficients.

[0096] Furthermore, the matching module 16 is also used to perform the following steps: perform cluster analysis on the historical claims results corresponding to K historical transportation tasks to identify high-frequency loss types in the claims results; establish risk-dominant labels based on the distribution ratio of high-frequency loss types in the K historical transportation tasks; use the risk-dominant labels to perform similar scenario filtering on the K historical transportation tasks, eliminate historical transportation tasks whose loss types deviate from the risk-dominant labels, and establish similar scenario matching results.

[0097] Furthermore, the matching module 16 is also used to perform the following steps: statistically analyze the historical compensation amount and loss coverage of different insurance types in the similar scenario matching results; establish insurance type loss compensation capability parameters based on historical compensation amount, and establish insurance type risk coverage capability parameters based on loss coverage; perform weighted fusion calculation on the loss compensation capability parameters and risk coverage capability parameters of each insurance type based on the dominant loss mode corresponding to the target cargo transportation task, and establish insurance type coverage contribution value; output the insurance matching result corresponding to the target transportation task according to the insurance type coverage contribution value sorting result.

[0098] In Embodiment 3, based on the same inventive concept as the machine learning-based cargo transportation insurance matching method in the foregoing embodiments, this application also provides a computer-readable storage medium storing a computer program, which, when executed, implements the steps of the machine learning-based cargo transportation insurance matching method described in any one of Embodiment 1.

[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A cargo transportation insurance matching method based on machine learning, characterized in that, The method includes: Read the transportation data of the target cargo transportation task, including cargo attribute data, transportation route data, carrier data, and historical transportation claims data; The transportation demand is analyzed from the cargo attribute data in the transportation data to establish cargo transportation demand characteristics that characterize the target cargo's transportation condition requirements during the transportation process; The system performs transportation capacity analysis on carrier data, and uses historical transportation records, transportation equipment information, and transportation performance data to establish transportation capacity characteristics that characterize the actual transportation capacity of carriers. Based on the characteristics of cargo transportation demand and carrying capacity, an adaptation deviation analysis is conducted to establish characteristics of transportation mismatch. By utilizing transportation route data, we can perform historical risk event correlation analysis on the corresponding transportation segments of the target transportation route, establish the risk amplification trend of transportation mismatch characteristics in the transportation process, and establish transportation risk characteristics. Based on transportation risk characteristics and historical transportation claims data, similar scenarios are matched, and the insurance coverage matching is performed using the similar scenario matching results to output the insurance matching results.

2. The machine learning-based cargo transportation insurance matching method as described in claim 1, characterized in that, Based on the characteristics of cargo transportation demand and carrying capacity, an adaptation deviation analysis is conducted to establish characteristics of transportation mismatch, including: Establish transportation assurance dimensions, which include temperature control range deviation, vibration power spectral density deviation, delivery timeliness deviation, loading and unloading operation standardization deviation, and packaging impact resistance margin deviation. Under the transportation guarantee dimension, the characteristics of cargo transportation demand and the characteristics of carrying capacity are calculated by dimension-by-dimensional difference to establish a multi-dimensional transportation mismatch vector, and the multi-dimensional transportation mismatch vector is output as the characteristic of the degree of transportation mismatch.

3. The machine learning-based cargo transportation insurance matching method as described in claim 2, characterized in that, Establish transportation risk characteristics, including: The transportation route data is parsed to extract N consecutive transportation segments along the target transportation route. Environmental entity attributes and historical risk labels are labeled for each transportation segment. The environmental entity attributes include the segment's terrain ruggedness, probability of seasonal strong winds or rainfall, and road surface grade. The historical risk labels include the baseline probability distribution of cargo damage, cargo discrepancies, and delay events in the segment. A spatiotemporal risk amplification inference network is established, wherein the spatiotemporal risk amplification inference network takes transportation segments as nodes and transportation segment connection relationships as edges, and includes a structural causal model of mismatch condition input interface; The environmental entity attributes and historical risk labels are concatenated and then input into the spatiotemporal risk amplification inference network to perform supervised training of network parameters. The multidimensional transport mismatch vector is loaded into the trained spatiotemporal risk amplification inference network, and forward inference is performed segment by segment along the path direction defined by the segment attribute sequence to establish a multidimensional risk time series feature curve that reflects the evolution of risk along the path. Multi-scale one-dimensional convolution and self-attention pooling operations are performed on the multi-dimensional risk time-series feature curve to extract the path global risk representation vector; Transportation risk features are established based on the global risk representation vector of the described path.

4. The machine learning-based cargo transportation insurance matching method as described in claim 3, characterized in that, Based on the global risk representation vector of the route, transportation risk features are established, including: Based on cargo attribute data, the inherent vulnerability parameters of the target cargo are analyzed, including impact brittleness value, temperature change tolerance gradient, and value density. The path global risk representation vector is multiplied by the inherent vulnerability parameter to generate a high-dimensional transportation risk feature tensor, which is then output as the transportation risk feature.

5. The machine learning-based cargo transportation insurance matching method as described in claim 1, characterized in that, Matching similar scenarios based on transportation risk characteristics and historical transportation claims data, including: The transportation risk features are flattened into a one-dimensional risk feature vector, and the one-dimensional historical risk feature vector corresponding to each historical transportation task is extracted simultaneously from the historical transportation claims data. Assign a learnable dynamic weight coefficient to each component of the one-dimensional risk feature vector; The weighted cosine similarity is calculated based on the weighted one-dimensional risk feature vector and the weighted one-dimensional historical risk feature vector, and K historical transportation tasks with similarity higher than a preset threshold are extracted as similar scene matching results.

6. The machine learning-based cargo transportation insurance matching method as described in claim 5, characterized in that, The configuration of dynamic weighting coefficients includes: Construct sample pairs by obtaining real insurance type selections and claims results from historical insurance matching records; With the goal of maximizing the correlation between matching similarity and insurance matching accuracy, the weights of risk features in each dimension are learned through gradient descent, and dynamic weight coefficients are established.

7. The machine learning-based cargo transportation insurance matching method as described in claim 5, characterized in that, Extract K historical transportation tasks with similarity higher than a preset threshold as similarity scene matching results, including: Cluster analysis is performed on the historical claims results corresponding to K historical transportation tasks to identify high-frequency loss types in the claims results; Establish risk-dominant labels based on the distribution of high-frequency loss types in the K historical transportation tasks; The risk-dominant label is used to perform similar scenario filtering on K historical transportation tasks, and historical transportation tasks whose loss type deviates from the risk-dominant label are removed to establish similar scenario matching results.

8. The machine learning-based cargo transportation insurance matching method as described in claim 7, characterized in that, Perform insurance coverage matching using similar scenario matching results, and output insurance matching results, including: Statistical analysis of historical payout amounts and loss coverage for different insurance types in similar scenario matching results; Establish parameters for the loss compensation capability of insurance products based on historical claims amounts, and establish parameters for the risk coverage capability of insurance products based on the scope of loss coverage. Based on the dominant loss mode corresponding to the target cargo transportation task, a weighted fusion calculation is performed on the loss compensation capability parameters and risk coverage capability parameters of each type of insurance to establish the insurance coverage contribution value. Output the insurance matching results corresponding to the target transportation task based on the sorting results of the insurance coverage contribution value.

9. A cargo transportation insurance matching system based on machine learning, characterized in that, The system is used to execute the machine learning-based cargo transportation insurance matching method according to any one of claims 1 to 8, the system comprising: The reading module is used to read the transportation data of the target cargo transportation task. The transportation data includes cargo attribute data, transportation route data, carrier data, and historical transportation claims data. The parsing module is used to analyze the cargo attribute data in the transportation data to determine transportation demand and establish cargo transportation demand characteristics that characterize the target cargo's transportation condition requirements during the transportation process. The execution module is used to perform transportation capacity analysis on the carrier's data, and to establish transportation capacity characteristics that represent the actual transportation capacity of the carrier by using historical transportation records, transportation equipment information, and transportation performance data. The analysis module is used to perform adaptation deviation analysis based on cargo transportation demand characteristics and carrying capacity characteristics, and to establish transportation mismatch characteristics. The risk module is used to perform historical risk event correlation analysis on the corresponding transportation segment of the target transportation route using transportation route data, establish the risk amplification trend of transportation mismatch characteristics in the transportation process, and establish transportation risk characteristics. The matching module is used to perform similar scenario matching based on transportation risk characteristics and historical transportation claims data, and to perform insurance coverage matching using the similar scenario matching results, and output the insurance matching results.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is used to execute the machine learning-based cargo transportation insurance matching method according to any one of claims 1 to 8.

Images