Integrated system of security Internet of Things and intelligent evaluation of insurance claim settlement

Abstract

The invention relates to the field of Internet of Things technology, data analysis and insurance claim settlement, in particular to a security Internet of Things and insurance claim settlement intelligent evaluation integrated system, which comprises a security state data acquisition module, a dynamic security portrait module, an accident traceability and responsibility intelligent division module and an automatic claim settlement evaluation module. According to the method, deep analysis of multi-source heterogeneous environment data and accurate representation of the safety state are achieved, the dynamic safety portrait module can capture the essential topological structure relation of the data by constructing the functions of topological feature extraction, state representation, multi-scale anomaly detection, risk prediction and the like, and high-precision recognition and risk prediction of gradient anomaly are achieved. After an accident occurs, a complete accident data chain can be constructed, the responsibility proportion of each party can be intelligently divided based on objective data, and the compensation amount can be automatically calculated, so that full-process automation from safety monitoring to insurance claim settlement is realized, the anomaly detection accuracy is greatly improved, and potential risks can be found in advance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things, data analysis and insurance claim, in particular to a safety Internet of Things and insurance claim intelligent evaluation integrated system for realizing full-process automation from environment state monitoring to insurance claim evaluation. BACKGROUND

[0002] With the development of Internet of Things technology, various types of sensing devices are widely used in industrial, commercial and residential environments, providing rich data sources for safety state monitoring. Traditional safety monitoring systems mainly focus on simple anomaly detection and alarm functions, and it is difficult to accurately characterize and predict complex and variable environment states. At the same time, the traditional insurance claim process is highly dependent on manual investigation and judgment, and has problems such as low efficiency, strong subjectivity and many disputes.

[0003] In the prior art, safety monitoring and insurance claim are usually operated as two independent systems, lacking effective information sharing and coordination mechanisms. Safety monitoring systems usually perform anomaly detection based on simple threshold judgments, which are difficult to adapt to dynamically changing environments; insurance claim systems lack the ability to analyze complete data chains before and after accidents, resulting in insufficient accuracy of responsibility division and loss assessment.

[0004] In addition, the anomaly detection methods in the prior art are mostly based on statistical methods and machine learning models, and have limited processing capacity for high-dimensional heterogeneous data, making it difficult to capture the topological structure relationship between data, resulting in low detection accuracy, especially for gradual abnormality recognition.

[0005] Therefore, there is an urgent need for a system that can integrate safety monitoring and insurance claim, improve anomaly detection and risk prediction capabilities through advanced data analysis technology, and support objective data-based responsibility division and claim evaluation. SUMMARY

[0006] The purpose of the present application is to provide a safety Internet of Things and insurance claim intelligent evaluation integrated system, aiming to solve the problems of separation of safety monitoring and insurance claim, low accuracy of anomaly detection and strong subjectivity of the claim process in the prior art.

[0007] The present application provides a safety Internet of Things and insurance claim intelligent evaluation integrated system, comprising:

[0008] a safety state data acquisition module for acquiring multi-source heterogeneous environment data, the multi-source heterogeneous environment data including video image data, device operating state data and multi-dimensional sensor data;

[0009] A dynamic security portrait module, in communication connection with the security state data collection module, configured to receive the multi-source heterogeneous environment data, construct a multi-dimensional data topology representation based on topology data analysis theory, and generate a security state evaluation result with persistent characteristics;

[0010] An accident traceability and responsibility intelligent division module, in communication connection with the dynamic security portrait module, configured to receive the security state evaluation result when detecting a security anomaly, analyze a cause chain of the accident, and determine a responsibility proportion of each related party;

[0011] An automatic claim assessment module, in communication connection with the security state data collection module and the accident traceability and responsibility intelligent division module, configured to automatically calculate an insurance compensation amount according to the multi-source heterogeneous environment data and the responsibility proportion.

[0012] Preferably, the security state data collection module comprises:

[0013] A monitoring node group configured to collect video image data, equipment running state data, and people flow data;

[0014] A sensor node group configured to collect temperature, humidity, illumination, smoke, CO concentration, noise, radiation, and equipment switch state data;

[0015] A data aggregation unit connected with the monitoring node group and the sensor node group, configured to perform preliminary format conversion and time stamp unification on the collected data to form a time-sequenced environment state data stream.

[0016] Preferably, the dynamic security portrait module comprises:

[0017] A data preprocessing unit configured to perform cleaning, standardization, and synchronization processing on the multi-source heterogeneous environment data, and convert all data into a standard format;

[0018] A topology feature extraction unit connected with the data preprocessing unit, configured to construct a multi-dimensional feature space, generate a multi-scale topology representation, and calculate a persistent homology group;

[0019] A state representation unit connected with the topology feature extraction unit, configured to construct a state transition network based on a persistent graph, and dynamically represent an environment security state;

[0020] An anomaly detection unit connected with the state representation unit, configured to calculate a topology distance between a current state and a normal template, detect an abnormal state, and perform positioning.

[0021] Preferably, the topology feature extraction unit comprises:

[0022] a distance function evaluation component configured to adaptively select an optimal distance function according to different sensor data characteristics;

[0023] a topological space construction component connected to the distance function evaluation component and configured to construct a simplicial complex under multiple threshold parameters;

[0024] a persistent homology calculation component connected to the topological space construction component and configured to calculate the life cycle of topological features of different dimensions;

[0025] a persistent diagram generation component connected to the persistent homology calculation component and configured to visualize the topological features as a two-dimensional scatter plot.

[0026] Preferably, the state characterization unit comprises:

[0027] a time series analysis component configured to construct a time series of the topological features, calculate topological entropy, and evaluate system stability;

[0028] a state discretization component connected to the time series analysis component and configured to divide a continuous topological feature space into a finite state region;

[0029] a transition matrix calculation component connected to the state discretization component and configured to count state transition frequencies and construct a Markov state transition matrix;

[0030] a reference update component connected to the transition matrix calculation component and configured to dynamically update a normal state topological template according to environmental changes and equipment aging.

[0031] Preferably, the anomaly detection unit comprises:

[0032] a multi-distance calculation component configured to calculate the Wasserstein distance and the bottleneck distance between the current persistent diagram and the normal template;

[0033] a multi-scale analysis component connected to the multi-distance calculation component and configured to detect topological anomalies at microscopic, mesoscopic, and macroscopic scales;

[0034] an anomaly positioning component connected to the multi-scale analysis component and configured to evaluate the contribution of each sensor to the anomaly and determine the spatial position of the anomaly source;

[0035] a risk prediction component connected to the anomaly positioning component and configured to analyze the evolution trend of the topological features and predict the future development trajectory of the system state.

[0036] Preferably, the accident tracing and responsibility intelligent division module comprises:

[0037] an accident data collection unit configured to collect data at the time of the accident and before and after the event, and construct an accident data chain;

[0038] The correlation analysis unit is connected with the accident data collection unit, and is configured to analyze the correlation and influence degree of each node in the accident chain.

[0039] The responsibility division unit is connected with the correlation analysis unit, and is configured to calculate the responsibility proportion of each party by using the Bayesian network and the decision tree technology.

[0040] The compensation calculation unit is connected with the responsibility division unit, and is configured to determine the compensation amount that should be borne by each party according to the responsibility proportion and the insurance contract terms.

[0041] Preferably, the automatic claim assessment module comprises:

[0042] The loss calculation unit is configured to divide the data into training samples and test samples, train the loss inference model, and calculate the numerical value and confidence of each possible loss.

[0043] The loss verification unit is connected with the loss calculation unit, and is configured to match the data of each monitoring node after the accident with the time node of the accident, and verify and correct the loss assessment.

[0044] The claim decision unit is connected with the loss verification unit, and is configured to generate a final claim scheme in combination with the responsibility division result and the loss assessment.

[0045] The dispute processing unit is connected with the claim decision unit, and is configured to transmit the data to the manual verification platform when the responsibility division is disputed.

[0046] Preferably, the system further comprises:

[0047] The risk index module is connected with the dynamic security portrait module and the security state data collection module, and is configured to construct a multi-level risk index based on environmental monitoring, equipment monitoring, personnel monitoring and security historical statistics, and generate a comprehensive risk score.

[0048] The insurance decision support module is connected with the risk index module and the automatic claim assessment module, and is configured to provide personalized insurance scheme suggestions for users based on the risk score and historical claim data.

[0049] Preferably, the system adopts a hierarchical deployment architecture, comprising:

[0050] The edge layer is deployed on the monitoring nodes and the sensor nodes, and is configured to perform data collection and preliminary feature extraction.

[0051] The server layer is deployed on the local server, and is configured to perform topology calculation, state network analysis and real-time anomaly detection.

[0052] The cloud layer is deployed on a cloud computing platform and is used for performing historical data analysis, model training and knowledge sedimentation.

[0053] Data is transmitted between layers through a secure communication channel to ensure real-time and security of data transmission.

[0054] The beneficial effects of the present application include:

[0055] 1. The whole process integration from safety state monitoring to insurance claim is realized, and the overall efficiency and synergy of the system are greatly improved;

[0056] 2. By introducing the topological data analysis theory, the processing capacity of high-dimensional heterogeneous data is improved, the essential characteristics of the environmental safety state can be accurately characterized, and the accuracy of anomaly detection is significantly improved, especially the identification ability of gradual abnormality is obviously improved;

[0057] 3. The accident traceability and responsibility division mechanism based on objective data is established, the subjective factors in the claim process are reduced, and the claim dispute is reduced;

[0058] 4. The early prediction and early warning ability of risk is realized, the potential risk can be found 10-30 minutes in advance by analyzing the evolution trend of topological characteristics, and a time window is provided for preventive measures;

[0059] 5. Through the hierarchical deployment architecture, the calculation efficiency and system complexity are balanced, so that the system has good scalability and adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 It is the overall architecture diagram of the safety Internet of Things and insurance claim intelligent evaluation integrated system of the present application.

[0061] Figure 2 It is the structural schematic diagram of the dynamic safety portrait module of the present application.

[0062] Figure 3 It is the work flow chart of the topological feature extraction unit of the present application.

[0063] Figure 4 It is the schematic diagram of the state transition network construction process of the present application.

[0064] Figure 5 It is the multi-scale anomaly detection schematic diagram of the present application.

[0065] Figure 6 It is the accident traceability and responsibility division flow chart of the present application.

[0066] Figure 7 It is the automatic claim evaluation flow chart of the present application. DETAILED DESCRIPTION

[0067] Please refer to Figures 1-7 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0068] like Figure 1 As shown, the integrated system for security IoT and intelligent insurance claims assessment provided by this invention mainly includes a security status data acquisition module 1, a dynamic security profile module 2, an accident tracing and intelligent liability division module 3, an automatic claims assessment module 4, a risk index module 5, and an insurance decision support module 6.

[0069] The safety status data acquisition module 1 collects multi-source heterogeneous environmental data, including video image data, equipment operating status data, and multi-dimensional sensor data. The dynamic safety profiling module 2 communicates with the safety status data acquisition module 1, receives the multi-source heterogeneous environmental data, constructs a multi-dimensional data topology representation based on topological data analysis theory, and generates a safety status assessment result with persistent characteristics. The accident tracing and intelligent liability allocation module 3 communicates with the dynamic safety profiling module 2. When a safety anomaly is detected, it receives the safety status assessment result, analyzes the cause chain of the accident, and determines the liability ratio of each relevant party. The automatic claims assessment module 4 communicates with both the safety status data acquisition module 1 and the accident tracing and intelligent liability allocation module 3, and automatically calculates the insurance compensation amount based on the multi-source heterogeneous environmental data and the liability ratio.

[0070] Preferably, the risk index module 5 is connected to the dynamic safety profile module 2 and the safety status data acquisition module 1, and constructs a multi-level risk index based on environmental monitoring, equipment monitoring, personnel monitoring, and historical safety statistics to generate a comprehensive risk score. The insurance decision support module 6 is connected to the risk index module 5 and the automatic claims assessment module 4, and provides users with personalized insurance plan suggestions based on the risk score and historical claims data.

[0071] Reference Figure 1 The safety status data acquisition module 1 includes a monitoring node group 11, a sensor node group 12, and a data aggregation unit 13.

[0072] The monitoring node group 11 is used to collect video image data, equipment operating status data, and people flow data. In one embodiment of the present invention, the monitoring node group 11 includes video surveillance equipment, intelligent electrical equipment, and an access control system. The video surveillance equipment collects video images with a resolution of 1080P and a frame rate of 25fps; the intelligent electrical equipment collects energy efficiency data such as voltage, current, and power, with a sampling frequency of 1Hz; the access control system collects personnel entry and exit records and identity authentication information.

[0073] Sensor node group 12 is used to collect data on temperature, humidity, light, smoke, CO concentration, noise, radiation, and equipment on / off status. In a specific implementation of this invention, the temperature and humidity sensors have a measurement range of -40℃ to 80℃ (temperature) and 0% to 100%RH (humidity), with accuracies of ±0.5℃ and ±3%RH, respectively; the detection thresholds for the smoke sensor and CO sensor are set to 10% and 30ppm of the standard concentration, respectively; the noise sensor has a measurement range of 30 to 120dB; and the radiation sensor is selected according to the application scenario, such as an ionizing radiation sensor or an electromagnetic radiation sensor.

[0074] The data aggregation unit 13 is connected to the monitoring node group 11 and the sensor node group 12, performing preliminary format conversion and timestamp unification on the collected data to form a time-series environmental status data stream. This invention employs a unified data format standard to ensure effective fusion and analysis of data from different sources. The timestamp precision is set to milliseconds to support high-precision event correlation analysis.

[0075] like Figure 2 As shown, the dynamic security profile module 2 is the core component of this system, including a data preprocessing unit 21, a topology feature extraction unit 22, a state representation unit 23, and an anomaly detection unit 24.

[0076] The data preprocessing unit 21 cleans, standardizes, and synchronizes the multi-source heterogeneous environmental data, transforming all data into a standard format. Specifically, the data cleaning process includes outlier detection and handling, missing value imputation, and signal denoising. For outlier detection, this invention employs a sliding window-based interquartile range (IQR) method, setting upper and lower limits as follows:

[0077] Upper limit=Q3+1.5×IQR,

[0078] Lower limit = Q1 - 1.5 × IQR

[0079] Where: Q1 is the first quartile of the data, that is, the value at the 25th percentile after sorting; Q3 is the third quartile of the data, that is, the value at the 75th percentile after sorting; IQR = Q3 - Q1 is the interquartile range, which represents a robust estimate of the dispersion of the data.

[0080] For handling missing values, different interpolation strategies are selected according to the data type: linear interpolation or mean interpolation is used for numerical data, mode replacement is used for categorical data, and interpolation based on historical patterns is used for time series data.

[0081] Data standardization employs an improved Min-Max normalization method, taking into account historical distribution and avoiding the influence of extreme values.

[0082] ,

[0083] Where: X is the original data value; X h ᵢ s This is a historical dataset containing historical observations over a sufficiently long period (typically 30 days). and These are the minimum and maximum values ​​of historical data, respectively. These are normalized data values, ranging from 0 to 1.

[0084] To ensure effective alignment of sensor data with different sampling rates, this system implements a time series alignment mechanism based on the principle of Dynamic Time Warping (DTW), which achieves uniform sampling rate while maintaining signal characteristics.

[0085] The topological feature extraction unit 22 is one of the innovations of this invention. Based on topological data analysis theory, it extracts essential features from multidimensional data that do not change with minor data variations. For example... Figure 3 As shown, the unit includes a distance function evaluation component 221, a topology space construction component 222, a persistent coherence calculation component 223, and a persistent graph generation component 224.

[0086] The distance function evaluation component 221 adaptively selects the optimal distance function based on the characteristics of different sensor data. The system maintains a distance function library, including Euclidean distance, Manhattan distance, Mahalanobis distance, and cosine similarity. For different types of sensor data, the system dynamically selects the most suitable distance function based on the data distribution characteristics and historical performance. For example, Euclidean distance is preferred for continuous numerical data such as temperature and humidity; Manhattan distance is preferred for discrete data such as equipment status; and cosine similarity is considered for high-dimensional feature vectors.

[0087] The topology space construction component 222 is connected to the distance function evaluation component 221 to construct a simplicial complex under multiple threshold parameters. Specifically, the system defines a distance threshold sequence. ,in A typical configuration is a series of geometrically increasing values ​​from the minimum distance to the maximum distance, such as... wait.

[0088] At each threshold Next, construct the Vietoris-Rips complex: if data points and Distance between Then, an edge is added to the complex to connect the two points. In this way, as the threshold increases, the simple complex will contain more and more edges and high-dimensional simplexes (such as triangles, tetrahedrons, etc.), thereby capturing the topological features of the data. To reduce computational complexity, the system employs sparsity techniques, prioritizing the retention of key topological features. In practical applications, this can reduce the computational complexity from... Reduce to ,in This represents the number of data points.

[0089] The persistent homology computation component 223 is connected to the topology space construction component 222 to compute the lifetime of topological features in different dimensions. In topology data analysis, the k-dimensional homology group... Representing the k-dimensional topological features in the data: Indicates connected components, Indicates a ring (or hole). This indicates a cavity, and so on.

[0090] By calculating the homology groups at different thresholds, the life and death of topological features can be tracked: the features at a certain threshold... The occurrence (“birth”) occurs at another, larger threshold. Disappearance ("death"). The persistence of a feature is defined as its lifetime length. Features with high persistence usually represent the essential structure of the data, while features with low persistence may just be noise.

[0091] The persistent graph generation component 224 is connected to the persistent coherence calculation component 223 to visualize topological features as a two-dimensional scatter plot. The persistent graph is a powerful tool for representing the persistence of topological features; the horizontal axis represents the birth threshold of a feature, and the vertical axis represents the death threshold. In the persistent graph, points farther from the diagonal represent features with stronger persistence.

[0092] In practical applications of this invention, persistence maps exhibit significant differences under different safety conditions. For example, persistence maps under normal conditions typically have regularly distributed feature points; fire precursor states may show specific patterns. Additional features; certain characteristics may precede equipment failure. Patterns such as premature disappearance of features can be identified. By recognizing these patterns, the system can accurately determine the safety status of the environment.

[0093] The state representation unit 23 is another innovation of this system. By tracking the temporal evolution of topological features, it constructs a dynamic state transition network, achieving accurate representation of the environmental safety state and capturing its changing trends. For example... Figure 4 As shown, the unit includes a time series analysis component 231, a state discretization component 232, a transition matrix calculation component 233, and a baseline update component 234.

[0094] The time series analysis component 231 constructs a time series of topological features, calculates topological entropy, and evaluates system stability. Specifically, the system employs a sliding window mechanism, with the window size dynamically adjusted according to the application scenario, typically ranging from 30 seconds to 5 minutes. For rapidly changing environments, such as industrial production lines, the window can be set to 30 seconds; for slower-changing environments, such as office buildings, the window can be set to 5 minutes.

[0095] The system converts the topological features of continuous time windows into a sequence of feature vectors and introduces the concept of topological entropy to quantify the complexity and uncertainty of the system state. The formula for calculating topological entropy is:

[0096] ,

[0097] in: Topological entropy represents the complexity and uncertainty of the system state; The normalized persistence of the i-th topological feature is represented by dividing the persistence of that feature by the sum of the persistence of all features. The total number of topological features; It is the natural logarithm; Indicates all The topological features are summed. Abrupt changes in topological entropy usually indicate critical transition points in the system state and are important indicators for early warning.

[0098] The state discretization component 232 is connected to the time series analysis component 231, dividing the continuous topological feature space into finite state regions. This invention employs a density-aware clustering method (such as the DBSCAN algorithm) to identify natural state boundaries. Compared to traditional uniform grid partitioning, this method better adapts to the natural distribution of data. The clustering parameters are set as follows: distance threshold. (Based on normalized feature space), minimum number of points .

[0099] Each identified state region is assigned a unique identifier and semantic label, such as normal operation, minor anomaly, or serious anomaly. These labels are determined based on historical data analysis and domain expert knowledge, making the system output more interpretable.

[0100] The transition matrix calculation component 233 is connected to the state discretization component 232 to statistically analyze the transition frequencies between states and construct the Markov state transition matrix. elements Indicates from state Transition to state The probability of:

[0101] ,

[0102] in: From state Transition to state The probability of; It is observed from historical data that the state Directly transition to state The number of times; From state The total number of transitions to any state, i.e., the number of states. The total number of times it appears; The total number of states is determined by the state discretization component. The system considers the transition patterns at different time scales and constructs short-term (within 10 minutes), medium-term (within 1 hour), and long-term (within 24 hours) transition matrices to capture system behavior patterns at different time scales.

[0103] The baseline update component 234 is connected to the transition matrix calculation component 233, and dynamically updates the normal state topology template according to environmental changes and equipment aging. The system identifies gradual changes caused by factors such as seasonal variations and equipment aging through long-term trend monitoring, and updates the template using a weighted fusion strategy.

[0104] ,

[0105] in: This is the updated template, representing the new standard for the normal state; The template before the update represents the standard of normal state accumulated over time; This represents the currently observed normal state; To update the weights, the value range is [0,1], typically set to 0.05-0.2 to ensure smooth template updates unaffected by short-term fluctuations. Smaller values... Values ​​(such as 0.05) are suitable for stable environments, while larger values ​​(such as 0.2) are suitable for environments with frequent changes.

[0106] To adapt to the differences in normal states under different operating modes, the system maintains a multimodal benchmark library, classifying benchmarks according to dimensions such as time (e.g., weekdays / rest days, daytime / nighttime), environmental conditions, and usage patterns, thereby achieving context-aware anomaly detection.

[0107] The anomaly detection unit 24 is the third innovation of this system. By analyzing the evolution of topological features, it predicts the future trajectory of the system state, achieving a shift from passive response to proactive early warning. For example... Figure 5 As shown, the unit includes a multi-distance calculation component 241, a multi-scale analysis component 242, an anomaly localization component 243, and a risk prediction component 244.

[0108] The multi-distance calculation component 241 calculates the Wasserstein distance and bottleneck distance between the current persistent graph and the normal template. The Wasserstein distance measures the optimal transmission cost between two persistent graphs, and is calculated using the following formula:

[0109] ,

[0110] in: For persistent graphs and Between Wasserstein distance; and There are two persistent graphs, each containing multiple points representing topological features; It is the set of all possible matching schemes, that is, from arrive The transmission plan; It is a metric space, which in this system is typically... (The coordinate space of points in the persistent graph); It is the Euclidean distance between point x and point y; This indicates that the minimum value is taken among all possible matching schemes; Represented in metric space Integrals on; This is the distance order, typically taken as 1 or 2. In practical implementations, to improve computational efficiency, the system employs an approximation algorithm, reducing the computational complexity from... Reduce to .

[0111] Bottleneck distance focuses on the differences in the most significant topological features:

[0112] ,

[0113] in: For persistent graphs and The bottleneck distance between them; From arrive The matching function; Indicates taking The maximum value of all points; It is a point Its matching point Maximum norm distance between them; This indicates that the minimum value is taken among all possible matching functions. The system comprehensively considers the results of these two distance metrics to improve detection robustness.

[0114] The multi-scale analysis component 242 is connected to the multi-distance calculation component 241 to detect topological anomalies at micro, meso, and macro scales. The micro scale focuses on deviations of individual data points or small areas, and is suitable for detecting sudden local anomalies; the meso scale identifies changes in local topological structures (such as rings and cavities), and can capture regional anomaly patterns; the macro scale monitors the shift in the overall distribution of topological features, and is suitable for discovering systemic anomalies.

[0115] The system sets different detection thresholds for different scales. For example, the Wasserstein distance threshold is set to 0.15 for the microscale, 0.25 for the mesoscale, and 0.35 for the macroscale. These thresholds were determined through extensive historical data analysis and expert experience, and can balance the detection rate and false alarm rate.

[0116] Anomaly localization component 243 is connected to multi-scale analysis component 242 to evaluate the contribution of each sensor to the anomaly and determine the spatial location of the anomaly source. The system employs a graph-based anomaly propagation analysis method to construct a sensor association network, calculate the anomaly propagation path, and identify the most likely anomaly source. The contribution calculation formula is as follows:

[0117] ,

[0118] in: It is a sensor The contribution level, with a value range of [0,1], indicates that the larger the value, the greater the contribution to the anomaly. Is it removing the sensor? The reduction in topological distance after data processing, i.e. , It is the topological distance calculated using all sensors. Is it removing the sensor? The topological distance is calculated afterward; It is the sum of contributions from all sensors, used for normalization; This represents the total number of sensors. Sensors with high contribution rates are identified as anomalies.

[0119] The risk prediction component 244 is connected to the anomaly localization component 243 to analyze the evolution trend of topological features and predict the future trajectory of the system state. The system constructs a topological feature evolution model and learns the evolution law of topological features based on historical data. First, the system extracts typical topological evolution trajectories from historical data and classifies and labels them according to the results (such as normal operation, gradual deterioration, sudden failure, etc.).

[0120] When a new abnormal pattern is identified, the system aligns the current trajectory with historical trajectories using the Dynamic Time Warping (DTW) algorithm and calculates the similarity:

[0121] ,

[0122] in: It is a trajectory and The dynamic time-normalized distance between them; and There are two trajectories, each consisting of topological features at a series of time points; It is the set of all possible alignment paths; Representing a path One of the alignment point pairs, which is about to The points and The Align the points; It is a point and The distance between them is usually expressed using Euclidean distance; This indicates that the minimum value is taken among all possible alignment paths.

[0123] The system selects the historical trajectory with the highest similarity as a reference and combines it with the current state transition network to make multi-step predictions: short-term predictions (within 5 minutes) are mainly based on the recently observed changing trends; medium-term predictions (within 30 minutes) combine the probabilistic model of the state transition network; and long-term predictions (within several hours) integrate the evolution patterns of historical similar trajectories.

[0124] Finally, the system comprehensively assesses the risk level based on the current degree of anomaly and the predicted evolution trend, thus achieving tiered early warning. The risk scoring formula is:

[0125] ,

[0126] in: It is a risk score, ranging from [0,1], with a higher value indicating a higher risk; It represents the degree of anomalousness of the current state, ranging from [0,1], and is calculated based on the topological distance normalized with respect to the normal template; It represents the severity of the predicted evolutionary trend, ranging from [0,1], based on the degree of similarity between the predicted trajectory and known dangerous states; It is an impact range assessment, with a range of [0,1], considering the areas and asset values ​​that may be affected by the anomaly; , and These are weight parameters that satisfy... The risk score is typically set to 0.3, 0.5, and 0.2 to emphasize the importance of predicting trends. The risk score is mapped to a multi-level warning level: Information (0-0.2), Caution (0.2-0.4), Warning (0.4-0.6), Emergency (0.6-0.8), and Critical (0.8-1.0).

[0127] like Figure 6 As shown, the accident tracing and intelligent liability division module 3 includes an accident data collection unit 31, a correlation analysis unit 32, a liability division unit 33, and a compensation calculation unit 34.

[0128] The accident data collection unit 31 collects data during the accident and before and after the event to construct an accident data chain. Specifically, the system utilizes sensor nodes and image monitoring equipment to collect all data before the accident (usually 5-30 minutes), during the accident, and after the accident (usually 1-4 hours), forming a complete accident data chain. This data includes sensor failure rate, environmental changes, equipment failure rate, abnormal personnel behavior, personnel authentication status, third-party verification, and accident loss information.

[0129] The association analysis unit 32 is connected to the accident data collection unit 31 to analyze the correlation and influence of each node in the accident chain. The system uses association rule mining algorithms (such as Apriori or FP-growth) to extract key attributes and event associations from the accident data chain. The support and confidence thresholds are set to 0.1 and 0.6, respectively, to ensure that significantly related attribute combinations are extracted.

[0130] In addition, the system employs causal network analysis to construct a causal relationship diagram of accident development based on time series correlation and domain knowledge, clarifying the causal chain between events and providing a basis for liability division.

[0131] The responsibility allocation unit 33 is connected to the correlation analysis unit 32, using Bayesian networks and decision tree techniques to calculate the responsibility ratio of each party. The Bayesian network model considers the conditional probability relationships of accident-related factors and calculates the degree of contribution of different entities to the accident. The decision tree model constructs a decision path for responsibility judgment based on historical cases and expert rules. The system integrates the results of these two models to provide a final responsibility allocation recommendation.

[0132] In practical applications, the system typically identifies the following responsible parties: facility owners, equipment suppliers, maintenance service providers, users, and third parties (such as natural factors). The proportion of responsibility for each party is calculated based on their contribution to the causal chain of the accident, accurate to a percentage.

[0133] The compensation calculation unit 34 is connected to the liability allocation unit 33, and determines the compensation amount to be borne by each party based on the liability ratio and the terms of the insurance contract. The system maintains an insurance contract terms database, and automatically retrieves and applies the corresponding compensation terms for different types of accidents and liability situations to calculate the final compensation amount.

[0134] like Figure 7As shown, the automatic claims assessment module 4 includes a loss calculation unit 41, a loss verification unit 42, a claims decision unit 43, and a dispute resolution unit 44.

[0135] The loss calculation unit 41 divides the data into training samples and test samples, trains the loss inference model, and calculates the numerical values ​​and confidence levels of various possible losses. In one embodiment of the invention, the system uses a random forest algorithm to train the loss inference model, considering the following key features: accident type, accident duration, affected area, equipment value, environmental conditions, and losses from similar historical cases. The model uses a 10-fold cross-validation method to evaluate its performance, achieving an accuracy of over 85%.

[0136] The loss verification unit 42 is connected to the loss calculation unit 41, matching the data from each monitoring node after the accident with the accident's timeline to verify and correct the loss assessment. The system collects actual status data after the accident, such as the extent of equipment damage and environmental recovery, and compares it with the prediction results of the inference model. When the difference exceeds 20%, the system adjusts the assessment results to ensure the accuracy of the loss assessment.

[0137] The claims decision-making unit 43 connects with the loss verification unit 42, and generates a final claims settlement plan by combining the liability allocation results and loss assessment. Based on the insurance contract terms, the system calculates the actual compensation amount for each liable party, taking into account factors such as deductibles, liability limits, and special conditions. The claims settlement plan includes detailed information such as the total compensation amount, the amount borne by each party, payment methods, and timing.

[0138] The dispute resolution unit 44 is connected to the claims decision-making unit 43. When there is a dispute over the apportionment of liability, the data is transmitted to the manual verification platform. When the system detects a highly complex or uncertain situation (such as a liability determination confidence level of less than 60%, or the involvement of new risk scenarios), it will automatically mark the case as requiring manual review and transmit all relevant data and analysis results to the manual verification platform for professional claims personnel to make the final judgment.

[0139] Risk index module 5 is connected to dynamic safety profile module 2 and safety status data acquisition module 1. Based on environmental monitoring, equipment monitoring, personnel monitoring, and historical safety statistics, it constructs a multi-level risk index and generates a comprehensive risk score. The risk index calculation formula is:

[0140] ,

[0141] in: This represents the overall risk index, ranging from [0,1], with higher values ​​indicating greater risk. Indicates the first The scores for each risk index level range from [0,1]; These are the corresponding weighting coefficients, satisfying... This represents the total number of risk index levels, typically 4, corresponding to environmental monitoring, equipment monitoring, personnel monitoring, and safety history statistics levels. The weight of each level is dynamically adjusted according to different application scenarios. For example, in an industrial environment, the weight of the equipment level may be higher (0.4), while in a commercial building, the weight of the personnel level may be higher (0.35).

[0142] The insurance decision support module 6 connects with the risk index module 5 and the automatic claims assessment module 4, providing users with personalized insurance plan suggestions based on risk scores and historical claims data. The system combines the user's risk preferences, protection needs, and budget constraints to match the most suitable plan from the insurance product library. The recommendation algorithm considers the following factors: the correlation between risk scores and premiums, the matching degree between historical claims patterns and protection needs, and user satisfaction feedback.

[0143] The system of this invention adopts a layered deployment architecture, including an edge layer, a server layer, and a cloud layer.

[0144] The edge layer is deployed on the monitoring and sensor node sides to perform data acquisition and preliminary feature extraction. Edge layer devices are typically embedded systems or industrial PCs, configured with ARM processors (such as Cortex-A72) and at least 2GB of RAM, running lightweight feature extraction algorithms to achieve preliminary data filtering and compression, reducing transmission bandwidth requirements.

[0145] The server layer is deployed on a local server and is used to perform topology calculations, stateful network analysis, and real-time anomaly detection. A typical configuration for the server layer is a multi-core x86 processor (such as an Intel Xeon), more than 32GB of memory, and GPU acceleration (such as an NVIDIA T4), which supports complex topology calculations and machine learning model execution, achieving millisecond-level response times.

[0146] The cloud layer is deployed on a cloud computing platform and is used to perform historical data analysis, model training, and knowledge accumulation. Cloud resources dynamically scale according to demand, supporting large-scale parallel computing, and are particularly suitable for handling long-term historical data analysis and complex model training.

[0147] Data is transmitted between layers via a secure communication channel using the TLS 1.3 encryption protocol to ensure data transmission security. The data transmission latency from the edge layer to the server layer is controlled within 100ms, and the latency from the server layer to the cloud is acceptable within 500ms, meeting the real-time requirements of different scenarios.

[0148] The integrated system of secure Internet of Things and intelligent insurance claims assessment of the present invention is applicable to a variety of application scenarios, including commercial buildings, industrial production environments and logistics supply chains.

[0149] In commercial building scenarios, the system deployment covers 50,000 square meters of commercial space, integrating over 1,000 sensor points and 200 video surveillance devices. Thirty computing nodes are deployed at the edge, and the central server cluster consists of eight high-performance servers. The implementation has yielded significant results: the system successfully issued warnings for 15 major security risks within one year, preventing potential losses of approximately 2 million yuan; claims processing time was reduced from an average of 7 days to 2 days; and insurance rates were reduced by 15%, saving customers substantial insurance costs.

[0150] In industrial production environments, the system monitors over 200 critical production devices, integrating data from industrial control systems and environmental monitoring. Customized topology feature templates are used for different types of equipment to improve the targeting of anomaly detection. After implementation, unexpected equipment downtime decreased by 35%, insurance rates decreased by 20%, and annual savings in maintenance and insurance costs exceeded 1 million yuan.

[0151] In logistics supply chain scenarios, the system is applied to monitor the entire cold chain logistics process to ensure cargo safety. The system integrates multi-dimensional sensor data such as temperature, humidity, and vibration to construct a topological feature model of the cargo's status. Implementation results show a 40% reduction in cargo damage rate, a 50% reduction in claims processing time, and a significant improvement in logistics service quality and customer satisfaction.

[0152] The adaptability and effectiveness of the system were verified through implementation in different scenarios, providing a successful example for the innovative integration of safety monitoring and insurance claims.

[0153] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. An integrated system for secure Internet of Things (IoT) and intelligent insurance claims assessment, characterized in that: include: The safety status data acquisition module is used to collect multi-source heterogeneous environment data, which includes video image data, equipment operating status data, and multi-dimensional sensor data. The dynamic security profiling module is communicatively connected to the security status data acquisition module. It is used to receive the multi-source heterogeneous environment data, construct a multi-dimensional data topology representation based on topology data analysis theory, and generate a security status assessment result with persistent characteristics. The accident tracing and intelligent responsibility allocation module is communicatively connected to the dynamic safety profiling module. When a safety anomaly is detected, it receives the safety status assessment results, analyzes the cause chain of the accident, and determines the responsibility ratio of each relevant party. The automatic claims assessment module is communicatively connected to the safety status data acquisition module and the accident tracing and liability intelligent division module, and is used to automatically calculate the insurance compensation amount based on the multi-source heterogeneous environmental data and the liability ratio.

2. The system according to claim 1, characterized in that, The security status data acquisition module includes: The monitoring node group is used to collect video image data, equipment operation status data, and people flow data; Sensor node groups are used to collect data on temperature, humidity, light, smoke, CO concentration, noise, radiation, and equipment on / off status. The data aggregation unit, connected to the monitoring node group and the sensor node group, is used to perform preliminary format conversion and timestamp unification on the collected data to form a time-series environmental status data stream.

3. The system according to claim 1, characterized in that, The dynamic security profile module includes: The data preprocessing unit is used to clean, standardize, and synchronize the multi-source heterogeneous environmental data, converting all data into a standard format. The topological feature extraction unit, connected to the data preprocessing unit, is used to construct a multi-dimensional feature space, generate multi-scale topological representations, and calculate persistent homology groups. The state representation unit, connected to the topology feature extraction unit, is used to construct a state transition network based on a persistent graph and dynamically represent the environmental security state. An anomaly detection unit, connected to the state characterization unit, is used to calculate the topological distance between the current state and the normal template, detect anomalies, and locate them.

4. The system according to claim 3, characterized in that, The topology feature extraction unit includes: A distance function evaluation component is used to adaptively select the optimal distance function based on the characteristics of different sensor data. A topology space construction component, connected to the distance function evaluation component, is used to construct a simple complex under multiple threshold parameters; A persistent coherence computing component, connected to the topology space construction component, is used to compute the lifecycle of topological features in different dimensions; A persistent graph generation component, connected to the persistent coherence calculation component, is used to visualize topological features as a two-dimensional scatter plot.

5. The system according to claim 3, characterized in that, The state representation unit includes: The time series analysis component is used to construct time series of topological features, calculate topological entropy, and evaluate system stability. A state discretization component, connected to the time series analysis component, is used to divide a continuous topological feature space into finite state regions; A transition matrix calculation component, connected to the state discretization component, is used to calculate the transition frequency between states and construct a Markov state transition matrix. The baseline update component, connected to the transition matrix calculation component, is used to dynamically update the normal state topology template according to environmental changes and equipment aging.

6. The system according to claim 3, characterized in that, The anomaly detection unit includes: The multi-distance calculation component is used to calculate the Wasserstein distance and bottleneck distance between the current persistent graph and the normal template; A multi-scale analysis component, connected to the multi-distance calculation component, is used to detect topological anomalies at micro, meso, and macro scales; An anomaly localization component, connected to the multi-scale analysis component, is used to evaluate the contribution of each sensor to the anomaly and determine the spatial location of the anomaly source. The risk prediction component, connected to the anomaly localization component, is used to analyze the evolution trend of topological features and predict the future development trajectory of the system state.

7. The system according to claim 1, characterized in that, The accident tracing and intelligent responsibility allocation module includes: The accident data collection unit is used to collect data during the accident and before and after the event to build an accident data chain. The correlation analysis unit, connected to the accident data collection unit, is used to analyze the correlation and impact of each node in the accident chain. The responsibility allocation unit, connected to the association analysis unit, is used to calculate the responsibility ratio of each party using Bayesian network and decision tree techniques. The compensation calculation unit, connected to the liability allocation unit, is used to determine the amount of compensation to be borne by each party based on the liability ratio and the terms of the insurance contract.

8. The system according to claim 1, characterized in that, The automatic claims assessment module includes: The loss calculation unit is used to divide the data into training samples and test samples, train the loss inference model, and calculate the numerical value and confidence level of various possible losses. The loss verification unit, connected to the loss calculation unit, is used to match the data of each monitoring node after the accident with the time node of the accident to verify and correct the loss assessment. The claims decision-making unit, connected to the loss verification unit, is used to generate a final claims settlement plan by combining the liability division results and loss assessment. The dispute resolution unit, connected to the claims decision-making unit, is used to transmit data to the manual verification platform when there is a dispute over the division of liability.

9. The system according to claim 1, characterized in that, The system also includes: The risk index module, connected to the dynamic safety profile module and the safety status data acquisition module, is used to construct a multi-level risk index based on environmental monitoring, equipment monitoring, personnel monitoring and safety history statistics, and generate a comprehensive risk score. The insurance decision support module, connected to the risk index module and the automatic claims assessment module, is used to provide users with personalized insurance plan suggestions based on risk scores and historical claims data.

10. The system according to claim 1, characterized in that, The system adopts a layered deployment architecture, including: The edge layer, deployed on the monitoring and sensor node sides, is used to perform data acquisition and preliminary feature extraction; The server layer, deployed on a local server, is used to perform topology calculations, stateful network analysis, and real-time anomaly detection. The cloud layer, deployed on a cloud computing platform, is used to perform historical data analysis, model training, and knowledge accumulation. Data is transmitted between different layers through a secure communication channel to ensure the real-time nature and security of data transmission.

Images