An internet-of-things-based monitoring method for movable property pledge

Abstract

The application discloses a kind of movables pledge monitoring methods based on Internet of Things, specifically related to financial risk control technical field, including the following steps: collection pledge individual multi-modal data, generate the value vector containing current individual value, value confidence and value change trend;Baseline is constructed based on group distribution, deviating individual is identified and marked to be confirmed abnormal event;According to signal continuity, node state and network log, distinguish abnormal source, filter false alarm and output real risk event;According to its risk level and confidence, adjust pledge rate and disposal intensity, and execute corresponding disposal operation;The application realizes individual value dynamic evaluation by generating value vector;By constructing group baseline and distinguishing abnormal source, filter false alarm event, improve risk identification accuracy;By dynamically adjusting pledge rate and disposal intensity according to risk level and confidence, realize flexible self-adapting risk control.

Description

Technical Field

[0001] This invention relates to the field of financial risk control technology, and more specifically, to a method for monitoring movable property pledges based on the Internet of Things. Background Technology

[0002] In movable asset pledge financing, continuous and effective monitoring of the pledged assets is a crucial aspect of ensuring the security of credit assets. Currently, existing technological solutions apply IoT technology to movable asset pledge monitoring. This involves deploying sensor nodes (such as electronic ear tags, weight sensors, and temperature and humidity sensors) and video surveillance equipment on or in the storage area of ​​the pledged assets to collect real-time location information, physiological parameters, movement trajectories, and environmental data. Some systems can trigger alarms based on preset single thresholds (such as location violations, signal interruption, or abnormal body temperature), or combine this with periodic manual inventory checks to verify the status of the pledged assets. Furthermore, some improved solutions introduce data fusion and remote monitoring platforms, enabling the uploading of sensor data to the cloud to assist lending institutions in risk assessment.

[0003] In practical operation, existing solutions still have some aspects that can be further optimized. On the one hand, sensor nodes may experience data anomalies or interruptions due to battery depletion, communication signal blockage, sensor drift, etc. These problems caused by equipment or communication links exhibit similar characteristics to actual risks such as the actual loss of collateral and value decay at the monitoring data level, easily leading to false alarms or missed alarms, increasing manual verification costs, and affecting risk control efficiency. On the other hand, existing solutions mostly use fixed thresholds or simple rules for anomaly judgment, which is difficult to adapt to the differences between different individual collateral (such as different breeds and different growth stages of livestock) and the dynamic changes in the value of collateral. Therefore, this paper proposes an IoT-based movable property pledge monitoring method to solve the above problems. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring movable property pledges based on the Internet of Things includes the following steps: By deploying IoT sensor nodes and environmental monitoring equipment on the pledged assets, multimodal perception data of each pledged entity is collected in real time, and the current value of each pledged entity and the trend of its change are dynamically calculated to generate a value vector containing the current value, value confidence, and value change trend. The value vectors are aggregated into a group value distribution, and a group behavior baseline is constructed based on the statistical characteristics of the group value distribution. The value vector of each pledged individual is compared with the group behavior baseline in real time to identify individuals that deviate from the group behavior baseline and mark the deviation as an anomaly to be confirmed. A list of anomalies to be confirmed containing the anomaly type, the degree of deviation, and the identifier of the associated individual is generated. For each unconfirmed anomaly in the list of confirmed anomalies, based on the signal continuity, sensor node status identifiers, and network communication logs in the multimodal sensing data, the source of the anomaly is distinguished as sensor node failure, communication interruption, or actual change in the state of the collateral. Anomalies originating from sensor node failure or communication interruption are marked as false alarms and filtered out. Only anomalies originating from actual changes in the state of the collateral are output as real risk events. Real risk events include risk level and risk confidence. Based on the risk level and risk confidence of the actual risk event, the risk control strategy parameters for monitoring movable property pledge are dynamically adjusted. The risk control strategy parameters include the pledge ratio and the intensity of disposal actions. Based on the adjusted intensity of the disposal action, the corresponding disposal operation of the pledged assets will be carried out.

[0005] In a preferred embodiment, the multimodal sensing data includes at least physiological parameters and behavioral data, with physiological parameters including body temperature and behavioral data including weight, daily exercise or food intake.

[0006] The intensity of the action is set in order of increasing risk level: only record reminder, reduce the pledge amount, partially freeze the pledged assets or freeze all the pledged assets, so as to achieve flexible adjustment that continuously matches the risk level.

[0007] In a preferred embodiment, the value vector is generated as follows: Extract at least one feature data that can be used for valuation from the multimodal perception data of the pledged individual. The feature data includes weight, body temperature, daily exercise or food intake. Calculate the current individual value: The current individual value is equal to the individual's benchmark value multiplied by the product of the correction factors corresponding to each feature data. Each correction factor is linearly determined according to the degree of deviation between the corresponding feature data and the standard value. The larger the deviation, the smaller the correction factor. Specifically, the correction factor is equal to one minus the ratio of the absolute value of the deviation to the maximum allowable deviation. The value confidence level is calculated based on the signal continuity and data integrity rate of the sensor nodes: the value confidence level is equal to one minus the ratio of the number of missing data points to the theoretical number of data points, and then multiplied by the preset signal attenuation coefficient; the preset signal attenuation coefficient is obtained by continuously collecting data from a standard signal source in an interference-free environment before system deployment, and statistically analyzing the average signal attenuation per unit time. The value change trend is calculated by fitting a linear trend to multiple historical value values ​​within the most recent time window: the value change trend is equal to the difference between the current value and the initial value of the window divided by the window duration. The calculated current individual value, value confidence level, and value change trend are combined into a value vector.

[0008] In a preferred embodiment, the process of constructing a group behavior baseline based on the statistical characteristics of group value distribution is as follows: Obtain the value vectors of all pledged individuals within the same regulatory batch. Each value vector contains three dimensions: current individual value, value confidence level, and value change trend. Then, perform benchmark normalization on the current individual value of each individual, that is, divide the current individual value by the benchmark value of that individual to obtain the relative value. Calculate the mean and standard deviation of all individuals in the relative value dimension, the mean and standard deviation in the value confidence dimension, and the mean and standard deviation in the value change trend dimension; The average values ​​of the three dimensions are combined to form the center vector, and the standard deviations of the three dimensions are combined to form the range vector. The center vector and the range vector together constitute the baseline of group behavior.

[0009] In a preferred embodiment, identifying individuals who deviate from the group's behavioral baseline refers to: For each pledged individual, its relative value, value confidence, and value change trend are compared with the average value of the corresponding dimension in the group behavior baseline. The ratio of the deviation value of each dimension to the standard deviation of that dimension is calculated. The square root of the sum of the squares of the three ratios is used to obtain the comprehensive deviation distance. Determine whether the overall deviation distance is greater than a preset deviation threshold, which is set to two; if it is greater, mark the individual as an individual who deviates from the group behavior baseline, and the degree of deviation is the value of the overall deviation distance.

[0010] In a preferred embodiment, the list of abnormal events to be confirmed is generated as follows: For individuals marked as deviating, the deviation multiples for each of the three dimensions—relative value, value confidence level, and value change trend—are calculated. The deviation multiple is the absolute value of the deviation in that dimension divided by the standard deviation of that dimension. The dimension with the largest deviation multiple among the three dimensions is selected as the primary deviation dimension; if multiple dimensions have the same largest deviation multiple, the primary deviation dimension is determined according to the priority order of value confidence, value change trend, and relative value. The type of abnormal event is determined based on the direction and attributes of the main deviation dimension: When the main deviation dimension is the value confidence level and its value is lower than the group average, the anomaly type is insufficient data credibility. When the main deviation dimension is the value confidence level and its value is higher than the group average, the anomaly event type is data abnormal stability. When the main deviation dimension is the value change trend and its value is lower than the group average, the abnormal event type is value growth stagnation; When the main deviation dimension is the value change trend and its value is higher than the group average, the abnormal event type is value growth that is too fast. When the main deviation dimension is relative value and its value is lower than the group average, the anomaly type is low value; When the main deviation dimension is a relative value and its value is higher than the group average, the anomaly type is value-biased. The abnormal event type, deviation degree, and individual identifier are combined into a single abnormal event to be confirmed. All abnormal events to be confirmed are then aggregated to generate a list of abnormal events to be confirmed.

[0011] In a preferred embodiment, the internal process of distinguishing the source of the abnormal event to be confirmed and outputting the actual risk event is as follows: For each unconfirmed anomaly, acquire the multimodal sensing data of its associated individuals, and extract signal continuity parameters, sensor node status identifiers, and network communication logs from them; the signal continuity parameters include the ratio of the most recent consecutive valid data frames to the theoretical number of frames; the sensor node status identifiers include the node battery voltage and sensor data self-test status; the network communication logs include the base station signal strength and packet loss rate. The source of anomalies is distinguished according to the preset hierarchical judgment rules. Then, the abnormal events that are judged to be sensor node failures or communication interruptions are marked as false alarm events and deleted from the list of abnormal events to be confirmed. Only the abnormal events that are judged to be changes in the actual state of the pledged assets are retained as real risk events. Calculate the risk level and risk confidence level of a real risk event: The risk level is determined by rounding up the ratio of the comprehensive deviation distance to the deviation threshold; the risk confidence level is equal to one minus the false alarm rate, where the false alarm rate is the ratio of the number of filtered false alarm events to the total number of abnormal events to be confirmed in the regulatory batch.

[0012] In a preferred embodiment, the preset hierarchy determination rule refers to: At the first level, if the battery voltage is lower than the operating voltage, the sensor node is considered to have failed. The second level is that if the battery voltage is not lower than the operating voltage but the signal continuity parameter is lower than the first threshold, then when the packet loss rate is higher than the second threshold or the base station signal strength is lower than the third threshold, it is determined that the communication is interrupted; otherwise, it is determined that the sensor node is faulty. At the third level, if the battery voltage is not lower than the operating voltage and the signal continuity parameter is not lower than the first threshold, but the sensor data self-test status shows that the measured value exceeds the calibration range, then the sensor node is determined to be faulty. At the fourth level, if none of the above conditions are met, it is determined that the actual state of the pledged property has changed.

[0013] In a preferred embodiment, the risk control strategy parameters for monitoring movable asset pledges are dynamically adjusted: Obtain the risk level and risk confidence level of a real risk event, where the risk level is an integer greater than or equal to one, and the risk confidence level is a real number greater than or equal to zero and less than or equal to one; The collateral ratio is adjusted based on the risk level and risk confidence level: the new collateral ratio is equal to the original collateral ratio multiplied by the adjustment factor, which is one minus the ratio of the current risk level to the maximum risk level, and then multiplied by the risk confidence level, where the maximum risk level is set to four. The intensity of the action is determined based on the risk level: when the risk level is one, the action intensity is to only record a reminder; when the risk level is two, the action intensity is to reduce the pledged amount; when the risk level is three, the action intensity is to partially freeze the pledged assets; when the risk level is four or higher, the action intensity is to freeze all the pledged assets. At the same time, the intensity of the action is adjusted according to the risk confidence level: if the risk confidence level is less than 0.5, the intensity of the action is lowered by one level; if the intensity of the action before the downgrade is already to only record a reminder, then no downgrade is made.

[0014] The technical effects and advantages of this invention are as follows: This invention utilizes IoT sensor nodes and environmental monitoring equipment deployed on the collateral to collect multimodal sensing data of each individual collateral in real time. It dynamically calculates the current individual value, value confidence level, and value change trend of each collateral, generating a value vector encompassing these three dimensions. This value vector not only reflects the current absolute value of the collateral but also characterizes data credibility through confidence level and reveals the dynamic trend of value through change trend. Compared to existing technologies that rely solely on a single fixed threshold (such as location violations or excessive body temperature) or simply record raw sensor values, this invention can adaptively handle differences in baseline value between different individuals (such as different breeds of livestock or different batches of goods), as well as continuous value changes caused by factors such as growth, health, and market fluctuations. This significantly improves the accuracy, real-time nature, and individual adaptability of collateral valuation, providing a more reliable data foundation for subsequent risk identification and flexible risk control.

[0015] This invention aggregates the value vectors of each individual pledgee into a group value distribution. Based on the statistical characteristics of the group value distribution, a group behavior baseline is constructed. Individuals deviating from the baseline are identified through real-time comparison and marked as anomalies to be confirmed. Furthermore, based on signal continuity, sensor node status identifiers, and network communication logs in multimodal sensing data, the source of anomalies is distinguished as sensor node failure, communication interruption, or changes in the actual state of the pledged assets. Anomalies originating from sensor node failure or communication interruption are marked as false alarms and filtered out, while only anomalies originating from changes in the actual state of the pledged assets are output as real risk events. This mechanism effectively overcomes the frequent false alarms caused by equipment or communication problems such as battery depletion, signal obstruction, and sensor drift in existing technologies, significantly reducing manual verification costs. Simultaneously, it avoids real risks (such as pledged asset loss or value decay) being overwhelmed by false alarms and missed, significantly improving the accuracy and reliability of risk identification.

[0016] This invention dynamically adjusts the risk control strategy parameters for movable asset pledge monitoring based on the risk level and risk confidence level of actual risk events. These parameters include the pledge ratio and the intensity of disposal actions, and ultimately, corresponding pledge disposal operations are executed based on the adjusted disposal action intensity. Specifically, the risk level reflects the severity of the pledged asset's deviation from the group baseline, and the risk confidence level characterizes the credibility of this judgment; together, they form a multi-dimensional basis for risk control decisions. When the risk confidence level is low, even if the risk level is high, the system will appropriately reduce the disposal intensity (e.g., downgrading "full freeze" to "partial freeze" or "reducing the pledge amount") to avoid excessive interference to the financing party due to possible false alarms. When both the risk confidence level and the risk level are high, timely and forceful disposal measures are taken to ensure the safety of funds. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of a movable property pledge monitoring method based on the Internet of Things in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Figure 1 The following examples were obtained: Example 1: A method for monitoring movable property pledge based on the Internet of Things, comprising the following steps: By deploying IoT sensor nodes and environmental monitoring equipment on the pledged assets, multimodal perception data of each pledged entity is collected in real time, and the current value of each pledged entity and the trend of its change are dynamically calculated to generate a value vector containing the current value, value confidence, and value change trend. The Internet of Things (IoT) sensor nodes are miniature intelligent electronic devices that are directly attached to each pledged individual (such as a cow or a shelf unit), and have unique identification, data collection, and wireless communication capabilities.

[0020] Typical equipment (taking live animal pledge as an example): Electronic ear tag: fixed to the animal's ear, with built-in temperature sensor and low-power wireless communication module, which can collect individual body temperature and unique identification.

[0021] Smart collar: Worn around the neck of livestock, it has built-in motion sensors (accelerometer, gyroscope) that can collect daily exercise volume, activity trajectory and rumination behavior.

[0022] Wearable weighing scale: Embedded in the standing area of ​​livestock (such as the weighing water trough pedal), it can collect weight data.

[0023] RFID tags: used for bulk commodities (such as steel coils, chemical drums) or packaged goods, collecting location and inbound / outbound information through readers.

[0024] Environmental monitoring equipment consists of fixed or mobile sensors deployed in warehouses, farms, or storage areas where the pledged assets are located, used to collect data on environmental conditions and group behavior of the pledged assets.

[0025] Typical equipment: High-definition network cameras: used for video surveillance of the distribution of pledged assets and identification of the number and location of individuals.

[0026] Base stations and signal repeaters: Wireless communication base stations deployed in the area are responsible for collecting data from sensor nodes and uploading it to the cloud, while also recording communication logs (such as signal strength and packet loss rate).

[0027] The value vectors are aggregated into a group value distribution, and a group behavior baseline is constructed based on the statistical characteristics of the group value distribution. The value vector of each pledged individual is compared with the group behavior baseline in real time to identify individuals that deviate from the group behavior baseline and mark the deviation as an anomaly to be confirmed. A list of anomalies to be confirmed containing the anomaly type, the degree of deviation, and the identifier of the associated individual is generated. For each unconfirmed anomaly in the list of confirmed anomalies, based on the signal continuity, sensor node status identifiers, and network communication logs in the multimodal sensing data, the source of the anomaly is distinguished as sensor node failure, communication interruption, or actual change in the state of the collateral. Anomalies originating from sensor node failure or communication interruption are marked as false alarms and filtered out. Only anomalies originating from actual changes in the state of the collateral are output as real risk events. Real risk events include risk level and risk confidence. Based on the risk level and risk confidence of the actual risk event, the risk control strategy parameters for monitoring movable property pledge are dynamically adjusted. The risk control strategy parameters include the pledge ratio and the intensity of disposal actions. Based on the adjusted intensity of the disposal action, the corresponding disposal operation of the pledged assets will be carried out.

[0028] The intensity of the action is set in order of increasing risk level: only record reminder, reduce the pledge amount, partially freeze the pledged assets or freeze all the pledged assets, so as to achieve flexible adjustment that continuously matches the risk level.

[0029] Multimodal perception data includes at least physiological parameters and behavioral data. Physiological parameters include body temperature, and behavioral data includes weight, daily activity level, or food intake. At least one feature data that can be used for valuation is extracted from the multimodal perception data of the pledged individual. This feature data includes weight, body temperature, daily activity level, or food intake. These features are extracted because weight directly determines the basic value of the pledged asset, body temperature reflects health status, and daily activity level and food intake jointly affect growth rate and value changes. Other feature data may be added as needed.

[0030] Calculating the current individual value: The current individual value equals the product of the individual's baseline value and the correction factors corresponding to each characteristic data point. Each correction factor is linearly determined based on the deviation of the corresponding characteristic data from the standard value; the larger the deviation, the smaller the correction factor. Specifically, the correction factor equals one minus the ratio of the absolute value of the deviation to the maximum permissible deviation. The baseline value is the initial valuation of the pledged individual under standard health conditions. For example, the baseline value of a healthy adult cow is 10,000 yuan, with a standard weight of 500 kg, a body temperature of 38.5 degrees Celsius, a daily activity level of 1,500 meters, and a feed intake of 8 kg. The maximum permissible deviation is set based on the normal fluctuation range of industry statistics. For example, the maximum permissible deviation for weight is 50 kg, and the maximum permissible deviation for body temperature is 0.5 degrees Celsius. When the actual weight is 470 kg, the absolute value of the deviation is 30 kg, and the correction factor equals one minus 30 divided by 50, which equals 0.4. Multiplying this factor by the baseline value yields the weight contribution portion of the current individual value.

[0031] The correction factors for all feature data are multiplied to obtain a comprehensive correction coefficient, which is then multiplied by the benchmark value to obtain the current individual value. The value confidence level is calculated based on the signal continuity and data integrity rate of the sensor nodes: the value confidence level equals one minus the ratio of missing data points to the theoretical number of data points collected, multiplied by a preset signal attenuation coefficient. The number of missing data points refers to the number of data points that should have been collected within a time window but were not actually collected. For example, if the theoretical number of data points collected is once per hour, or 24 times per day, and six data points are missing, the missing ratio is 0.25. One minus this ratio equals 0.75.

[0032] The preset signal attenuation coefficient is obtained as follows: Before system deployment, data from a standard signal source is continuously collected in an interference-free environment, and the average signal attenuation per unit time is calculated. For example, if a standard signal source is placed in an open area, the signal strength is -70 dB at 100 meters from the receiver and -90 dB at 200 meters. The average attenuation is -10 dB per 100 meters. This average value is then fixed into the system as the signal attenuation coefficient.

[0033] When the signal attenuation coefficient is 0.9, the value confidence level equals 0.75 multiplied by 0.9, which equals 0.675, indicating that the confidence level of the individual value data is 67.5%. The value change trend is calculated by fitting a linear trend to multiple historical value values ​​within the most recent time window: the value change trend equals the difference between the current value and the initial value of the window divided by the window duration. The length of the time window is set according to the value fluctuation cycle of the pledged asset; for example, for cattle with a long growth cycle, the window is set to seven days, and for fast-growing pigs, the window is set to three days.

[0034] If the current value is 8,000 yuan, and the initial value seven days ago was 7,500 yuan, the difference is 500 yuan. Dividing by seven days gives approximately 71.4 yuan per day, indicating that the value is increasing at a rate of approximately 71.4 yuan per day. The calculated current individual value, value confidence level, and value change trend are combined into a value vector. The value vector is an array containing three values, such as [8,000 yuan, 0.675, 71.4 yuan per day], for use in subsequent group analysis.

[0035] The process of constructing a baseline of group behavior based on the statistical characteristics of group value distribution is as follows: The system obtains the value vectors of all pledged individuals within the same regulatory batch. Each value vector includes three dimensions: current individual value, value confidence level, and value change trend. The current individual value of each individual is then normalized by dividing it by its benchmark value to obtain the relative value. A regulatory batch refers to a group of pledged individuals undergoing pledge monitoring at the same time and place. For example, consider fifty cows in the same pen. Different individuals may have different benchmark values; for instance, an adult cow might have a benchmark value of 10,000 yuan, while a calf might have a benchmark value of 5,000 yuan. Directly comparing current individual values ​​would be unfair. Normalization involves dividing the current individual value by its benchmark value to obtain the relative value. For example, a cow with a benchmark value of 10,000 yuan might currently have a value of 9,000 yuan, resulting in a relative value of 0.9; another cow with a benchmark value of 5,000 yuan might currently have a value of 4,800 yuan, resulting in a relative value of 0.96. Relative values ​​allow for comparison of individuals with different benchmarks.

[0036] Calculate the mean and standard deviation of all individuals in the relative value dimension, the mean and standard deviation in the value confidence dimension, and the mean and standard deviation in the value change trend dimension. The mean is the sum of all individual values ​​in that dimension divided by the total number of individuals. The standard deviation represents the dispersion among individuals in that dimension; a larger standard deviation indicates greater individual differences. For example, if the relative values ​​of fifty cows are 0.9, 0.95, and 1.0, the mean is 0.94 and the standard deviation is 0.05. The mean and standard deviation of the value confidence are calculated similarly; for example, if the average confidence is 0.8, the standard deviation is 0.1. The average value of the value change trend is 50 yuan per day, and the standard deviation is 20 yuan per day.

[0037] The average values ​​of the three dimensions are combined to form the central vector, and the standard deviations of the three dimensions are combined to form the range vector. The central vector and the range vector together constitute the baseline of group behavior. The central vector is an array of three average values, such as [relative value average 0.94, value confidence average 0.8, value change trend average 50 yuan per day]. The range vector is an array of three standard deviations, such as [relative value standard deviation 0.05, value confidence standard deviation 0.1, value change trend standard deviation 20 yuan per day].

[0038] The baseline for group behavior uses a center vector to describe the normal level of the group and a range vector to describe the normal fluctuation range of the group. For example, if the relative value of a cow fluctuates within 0.05 of 0.94, it is normal. If it exceeds this range, there may be a problem. In this way, the baseline does not need to be preset with a fixed threshold, but is dynamically determined according to the actual distribution of the current group, adapting to the differences between different batches and different breeds.

[0039] Identifying individuals who deviate from the group's behavioral baseline refers to: For each pledged individual, its relative value, value confidence level, and value change trend are compared with the average value of the corresponding dimension in the group behavior baseline. The ratio of the deviation value of each dimension to the standard deviation of that dimension is calculated. The square root of the sum of the squares of the three ratios is used to obtain the comprehensive deviation distance.

[0040] Each dimension has its own mean and standard deviation. Directly comparing deviation values ​​is affected by the different dimensions' units. For example, a relative value might be a fraction of a cent, while a value change trend might be tens of yuan per day; the units are different and cannot be directly added. Dividing the deviation value of each dimension by its standard deviation eliminates the dimension, resulting in a unitless value indicating "how many standard deviations" the value has deviated from. For example, if the relative value of a cow is 1.0, the group average relative value is 0.94, the standard deviation is 0.05, and the deviation is 0.06, dividing by the standard deviation gives 1.2 standard deviations. If the confidence level of the same cow's value is 0.5, the group average confidence level is 0.8, the standard deviation is 0.1, and the deviation is -0.3, dividing by the standard deviation gives -3 standard deviations. The value change trend is 10 yuan per day, the group average is 50 yuan per day, the standard deviation is 20 yuan per day, and the deviation value is -40. Dividing by the standard deviation gives -2 standard deviations, and the three ratios are 1.2, -3.0, and -2.0. Squaring them gives 1.44, 9.0, and 4.0. Adding them together gives 14.44, and taking the square root gives 3.8. This is the overall deviation distance. The larger the overall deviation distance, the farther the individual deviates from the group in all dimensions.

[0041] The algorithm determines whether the overall deviation distance exceeds a preset deviation threshold of two. If it does, the individual is marked as deviating from the group's behavioral baseline, and the degree of deviation is the value of the overall deviation distance. The reason for setting the deviation threshold to two is that, under a normal distribution, the probability of a value in one dimension deviating from the mean by more than two standard deviations is about five percent, and the probability of exceeding two after combining three dimensions is even lower, about two to three percent. Using two as a threshold can filter out about two to three percent of the most abnormal individuals. This proportion will not generate too many false alarms, nor will it miss any real anomalies. For example, if the overall deviation distance calculated above is 3.8, which is greater than two, then this cow is marked as a deviating individual, and the degree of deviation is 3.8. If another cow's overall deviation distance is 1.5, which is less than two, then it will not be marked and is considered to be within the normal fluctuation range.

[0042] The list of pending exception events is generated as follows: For individuals marked as deviating, the deviation multiples for each of the three dimensions—relative value, confidence level, and trend of value change—are calculated. The deviation multiple is the absolute value of the deviation in that dimension divided by the standard deviation of that dimension. The deviation multiple indicates the degree to which the individual deviates from the group average in a particular dimension; a larger value indicates a more severe deviation. For example, the cow in the previous section had a relative value deviation of 0.06 and a standard deviation of 0.05, resulting in a deviation multiple of 1.2; a confidence level deviation of 0.3 and a standard deviation of 0.1, resulting in a deviation multiple of 3.0; and a trend of value deviation of 40 and a standard deviation of 20, resulting in a deviation multiple of 2.0.

[0043] The dimension with the largest deviation multiple is selected as the primary deviation dimension from the three dimensions. If multiple dimensions have the same largest deviation multiple, a single primary deviation dimension is determined according to the priority order of value confidence level, value change trend, and relative value. The dimension with the largest deviation multiple is selected because it is the primary reason for an individual's deviation from the group, and subsequent anomalous event types should reflect this primary reason. For example, in the above example, the deviation multiples are 1.2, 3.0, and 2.0, with the largest being 3.0 for value confidence level; therefore, the primary deviation dimension is value confidence level. If both value confidence level and value change trend have the same largest deviation multiple of 3.0, value confidence level is selected as the primary deviation dimension according to priority, because confidence level anomalies are more likely to cause false alarms than trend anomalies and should be handled first.

[0044] The type of abnormal event is determined based on the direction and attributes of the main deviation dimension: When the main deviation dimension is the value confidence level and its value is lower than the population average, the anomaly type is insufficient data credibility. For example, the value confidence level of the cow mentioned above is 0.5, while the population average is 0.8, which is lower than the average, therefore the type is insufficient data credibility. This indicates that the individual's sensor data may be significantly lost or the signal may be severely attenuated, leading to unreliable value assessment.

[0045] When the main deviation dimension is the value confidence level and its value is higher than the population average, the anomaly type is data anomalously stable. For example, if a cow's value confidence level is 0.99 and the population average is 0.8, the deviation is 1.9. An excessively high confidence level implies almost no missing sensor data and excellent signal strength, but this is rare in real-world farming environments and may suggest that the sensor is fixed or the data is falsified; therefore, it is marked as data anomalously stable.

[0046] When the primary deviation dimension is the value change trend and its value is lower than the group average, the anomaly type is value stagnation. For example, if a cow's value change trend is five yuan per day, while the group average is fifty yuan per day, the value is lower than the average, the deviation multiple is 2.25, and the type is value stagnation. This indicates that the individual is growing slowly or losing weight, and the value of the pledged asset may decrease.

[0047] When the primary deviation dimension is the value change trend and its value is higher than the group average, the anomaly type is "value growth too rapid." For example, if the value change trend of a cow is 120 yuan per day, while the group average is 50 yuan per day, the cow is higher than the average, with a deviation factor of 3.5, and the type is "value growth too rapid." This could be due to measurement error or abnormal weight gain (such as water injection), which also constitutes a risk.

[0048] When the primary deviation dimension is relative value and its value is lower than the group average, the anomaly type is undervalued. For example, if a cow's relative value is 0.7 and the group average is 0.94, it is lower than the average, with a deviation factor of 4.8, and the type is undervalued. This indicates that the individual's current value is significantly lower than the normal level for its batch.

[0049] When the main deviation dimension is relative value and its value is higher than the group average, the anomaly type is "overvalued". For example, if a cow's relative value is 1.2 and the group average is 0.94, it is higher than the average, with a deviation factor of 5.2, and the type is "overvalued". This could be due to assessment model bias or an exceptionally high individual value, but it still requires attention.

[0050] The anomaly type, deviation degree, and individual identifier are combined into a single pending anomaly event. All pending anomalies are then compiled into a list. For example, the cow with low value confidence, identified as cow number A0123, with a deviation degree of 3.8, and an anomaly type of insufficient data confidence, constitutes a single pending anomaly event. If five cows are tagged within the same regulatory batch, five pending anomalies will be generated and added to the list for further source differentiation and processing.

[0051] The internal process for distinguishing the source of an anomaly to be confirmed and outputting the actual risk event is as follows: For each unconfirmed anomaly, multimodal sensing data of its associated individuals is acquired, from which signal continuity parameters, sensor node status identifiers, and network communication logs are extracted. Signal continuity parameters include the ratio of the most recent consecutive valid data frames to the theoretical number of frames. For example, if the theoretical number of frames is one data upload per hour, 24 times a day, but only 10 consecutive valid data frames have been received recently, then the ratio is 10 divided by 24, approximately 0.42. The smaller this ratio, the more frequent the data interruptions.

[0052] The sensor node status indicators include the node battery voltage and sensor data self-test status. A normal battery voltage is 3.7 volts; a voltage below 3.3 volts indicates insufficient power. The sensor data self-test status is the node's internal self-check result regarding whether measured values ​​are within a reasonable range. For example, if a body temperature sensor detects a measured value exceeding the 0-50 degree range during its self-test, it will be marked as abnormal.

[0053] Network communication logs include base station signal strength and packet loss rate. Base station signal strength is expressed in dBm; for example, -70 dBm is good, and -110 dBm is very poor. Packet loss rate refers to the proportion of transmitted data packets that are lost. For example, if 15 out of 100 packets are sent, the packet loss rate is 15%.

[0054] The system distinguishes the sources of anomalies based on preset hierarchical judgment rules. Events deemed to be sensor node failures or communication interruptions are then marked as false alarms and removed from the list of pending anomalies. Only events deemed to be changes in the actual state of the pledged assets are retained as genuine risk events. The preset hierarchical judgment rules will be explained in detail later. Here, the events are judged sequentially according to the rules. For example, if a cow is marked as a pending anomaly, and its signal continuity ratio is 0.3 and battery voltage is 3.2 volts (below the operating voltage), then according to the first level of rule, it is directly judged as a sensor node failure, and this event is marked as a false alarm and removed from the list. For another cow, the battery voltage is 3.6 volts (normal), the signal continuity ratio is 0.25, the packet loss rate is 30% (above the second threshold), and the base station signal strength is -115 dBm (below the third threshold). Therefore, according to the second level of rule, it is judged as a communication interruption, and similarly marked as a false alarm and removed. For example, with the third cow, the battery voltage is normal and the signal continuity ratio is 0.9 (normal), but the sensor data self-test status shows that the body temperature measurement value is outside the calibration range. According to the third level of the rule, this is judged as a sensor node failure and marked as a false alarm event.

[0055] Only at the fourth level, when none of the above conditions are met, is it determined that the actual state of the pledged property has changed. For example, if the battery voltage is normal, the signal is continuous and normal, the self-test is normal, and the communication log is normal, but the overall deviation is large, it indicates that the cow itself is really sick or has escaped, and it is retained as a real risk event.

[0056] The risk level and risk confidence level of a real risk event are calculated. The risk level is determined by rounding up the ratio of the comprehensive deviation distance to the deviation threshold. If the deviation threshold is set to two, and the comprehensive deviation distance is, for example, 3.8 as calculated previously, the ratio equals 3.8 divided by 2, which equals 1.9. Rounding up gives 2, so the risk level is two. If the comprehensive deviation distance is 5.1, the ratio is 2.55, rounding up gives 3, and the risk level is three. Rounding up means that if the decimal part is greater than zero, it is rounded up to the next integer to ensure that the risk level is at least one.

[0057] The risk confidence level equals one minus the false alarm rate, where the false alarm rate is the ratio of the number of filtered false alarm events to the total number of anomaly events to be confirmed within the regulatory batch. For example, if there are ten anomaly events to be confirmed within the regulatory batch, and after source differentiation, seven are determined to be sensor node failures or communication interruptions (i.e., false alarm events), and three are determined to be changes in the actual state of the pledged assets (real risk events), then the false alarm rate is seven divided by ten equals 0.7, and the risk confidence level equals one minus 0.7 equals 0.3.

[0058] A lower risk confidence level indicates that the judgment of a real risk event is less reliable, because most anomalies are false alarms. Conversely, if the false alarm rate is very low, such as only one false alarm, then the risk confidence level is 0.9, indicating that the judgment of a real risk event is very reliable. Ultimately, the risk level and risk confidence level outputs are used as additional attributes of real risk events for subsequent adjustment of risk control strategy parameters.

[0059] The preset hierarchical judgment rule refers to the following: At the first level, if the battery voltage is lower than the operating voltage, the sensor node is judged to have failed. The operating voltage refers to the minimum voltage value required for the normal operation of the sensor node; for example, the operating voltage of a common low-power IoT node is 3.3 volts. This threshold is set based on the undervoltage protection point specified in the datasheet of the chip inside the sensor node. When the voltage is lower than this value, both radio frequency communication and sensor measurement cannot guarantee accuracy, and the node enters a sleep or reset state. For example, the operating voltage range of a certain model of electronic ear tag is 3.0 volts to 3.7 volts. Using 3.3 volts as the operating voltage threshold allows for a 0.3 volt buffer zone to avoid misjudgments due to instantaneous voltage drops.

[0060] The second level determines whether communication is interrupted if the battery voltage is not lower than the operating voltage but the signal continuity parameter is lower than the first threshold. Communication is considered interrupted when the packet loss rate exceeds the second threshold or the base station signal strength is lower than the third threshold; otherwise, the sensor node is considered to have failed. The first threshold is the lower limit of the signal continuity parameter, set based on the communication protocol's requirement that the percentage of consecutive valid frames must be at least 50% to maintain a reliable link. For example, with a theoretical sampling frequency of once per hour (24 times per day), if the number of consecutive valid data frames is less than 12 (a ratio of 0.5), it indicates a severe interruption in the data flow. This threshold was determined through field testing: in environments with good signal strength, the signal continuity ratio is typically greater than 0.9; when the ratio is lower than 0.5, the positioning or tracking function is essentially unusable.

[0061] The second threshold is the upper limit of the packet loss rate, set based on the packet loss rate requirements for reliable data transmission in wireless communication industry standards, typically not exceeding 20%. When the packet loss rate exceeds 20%, the retransmission mechanism will significantly increase power consumption and latency, leading to data loss. For example, in actual tests in open areas, a packet loss rate below 5% is considered excellent, 10% to 20% is acceptable, and anything exceeding 20% ​​is deemed unacceptable for communication quality.

[0062] The third threshold is the lower limit of the base station signal strength, set based on the receiver sensitivity index, and is usually taken as -100 dBm. When the signal strength is below -100 dBm, the error rate of data packets increases sharply, and the connection becomes extremely unstable. For example, if the receiver sensitivity of a base station is -105 dBm, taking -100 dBm as the threshold can leave a 5 dB margin to avoid frequent handovers in edge states.

[0063] The third level: If the battery voltage is not lower than the operating voltage and the signal continuity parameter is not lower than the first threshold, but the sensor data self-test status shows that the measured value exceeds the calibration range, then the sensor node is determined to be faulty. The calibration range in the sensor data self-test status is set according to the sensor's factory calibration parameters. For example, the measurement range of a body temperature sensor is 35 degrees Celsius to 42 degrees Celsius; exceeding this range is considered a sensor failure. This threshold does not need to be set separately; the range boundary provided by the manufacturer is used directly. The fourth level: If none of the above conditions are met, then the actual state of the pledged item has changed.

[0064] Dynamically adjust the risk control strategy parameters for movable asset pledge monitoring: Obtain the risk level and risk confidence level of real risk events. The risk level is an integer greater than or equal to one, and the risk confidence level is a real number greater than or equal to zero and less than or equal to one. The risk level is calculated; for example, in the previous example, the comprehensive deviation distance of 3.8 divided by the deviation threshold of 2 equals 1.9, which is rounded up to 2, so the risk level is 2. A higher risk level indicates a more severe anomaly. The risk confidence level is equal to 1 minus the false alarm rate. For example, when the false alarm rate is 0.7, the confidence level is 0.3, indicating that the judgment of this real risk event has only a 30% confidence level.

[0065] The loan-to-value ratio (LTV) is adjusted based on risk level and risk confidence level: the new LTV equals the original LTV multiplied by an adjustment factor. The adjustment factor is one minus the ratio of the current risk level to the maximum risk level, then multiplied by the risk confidence level. The maximum risk level is set to four. The maximum risk level is set to four because there are four levels of action intensity (record only, reduction of pledged amount, partial freezing of pledged assets, and full freezing of pledged assets), with risk level four corresponding to the highest action intensity. When the risk level is four, the ratio of the current risk level to the maximum risk level is one. One minus this ratio equals zero, the adjustment factor is zero, and the new LTV is reduced to zero, indicating that no new loans will be issued.

[0066] For example: The original loan-to-value ratio is 70%, the current risk level is 2, and the risk confidence level is 0.8. The ratio of the current risk level to the maximum risk level is 2 divided by 4, which equals 0.5. 1 minus 0.5 equals 0.5. The adjustment factor is 0.5 multiplied by 0.8, which equals 0.4. The new loan-to-value ratio is 70% multiplied by 0.4, which equals 28%. This means that due to the increased risk and a more reliable assessment, the loan-to-value ratio has been reduced from 70% to 28%, meaning the borrower can only borrow less than half the original amount.

[0067] If the risk confidence level is very low, such as 0.3, the adjustment factor is 0.5 multiplied by 0.3, which equals 0.15. The new collateral ratio is 70% multiplied by 0.15, which equals 10.5%. A low risk confidence level indicates that the risk assessment is not very reliable, so the adjustment range is smaller, and the collateral ratio is lowered less to avoid excessive penalties for the financing party due to misreporting.

[0068] The intensity of the action is determined based on the risk level: when the risk level is one, the action intensity is to only record a reminder; when the risk level is two, the action intensity is to reduce the pledged amount; when the risk level is three, the action intensity is to partially freeze the pledged assets; when the risk level is four or higher, the action intensity is to freeze all the pledged assets.

[0069] The four risk levels are based on an increasing degree of risk severity. Risk level one indicates a slight deviation, requiring only record-keeping for future reference and not interfering with normal operations. Risk level two indicates moderate risk, requiring a reduction in the pledged amount to decrease financial exposure. Risk level three indicates high risk, with partial freezing of pledged assets (e.g., prohibiting the sale of 50% of the goods). Risk level four and above indicates extremely severe risk, with all assets frozen and any outflow or transfer prohibited.

[0070] Simultaneously, the intensity of the handling action is adjusted based on the risk confidence level: if the risk confidence level is less than 0.5, the intensity of the handling action is downgraded by one level; if the initial handling action intensity was already only recording a reminder, then no downgrade is made. The basis for this adjustment is that when the risk confidence level is less than 0.5, it indicates that there is a greater than 50% probability that the judgment of this real risk event is a false alarm, therefore a more lenient handling is implemented to avoid unnecessarily harming the financing party. For example, if the risk level of a real risk event is three, and partial freezing of pledged assets should be implemented, but the risk confidence level is only 0.3, less than 0.5, then the intensity of the handling action is downgraded by one level to reduce the pledged amount.

[0071] If the original risk level was 2, and the action intensity was to reduce the pledged amount, downgrading it by one level would change it to only recording a warning. If the original risk level was 1, and the action intensity was already only recording a warning, it could not be downgraded further, so it remained unchanged. For example, if the risk level was 4, all pledged assets should have been frozen, but the risk confidence level was 0.4, downgrading it by one level would change it to partially freezing the pledged assets. If the risk level was 4 and the confidence level was 0.6 (greater than or equal to 0.5), then no downgrading would be made, and all assets would still be frozen. This approach provides a baseline response based on the risk level, while also allowing for flexible adjustments based on the reliability of the assessment, avoiding overreactions to unreliable risk signals.

[0072] The internal process of executing the corresponding collateral disposal operation based on the adjusted disposal action intensity is as follows: obtain the determined disposal action intensity. When the disposal action intensity is to record only a reminder, the system writes a record in the monitoring log, which includes the event occurrence time, the associated collateral individual identifier, the abnormal event type, the risk level, and the risk confidence level. The system does not change the collateral status or send any control instructions.

[0073] When the disposal action is to reduce the pledged amount, the system recalculates the available loan amount for the borrower based on the new pledge ratio, sends a limit adjustment notification to the borrower, and updates the pledge status table internally, but does not restrict the physical movement of the pledged assets.

[0074] When the action intensity is to partially freeze the pledged collateral, the system sends a freeze instruction to the supervisor's electronic access control system or electronic fence system. The instruction includes the percentage that can be moved (e.g., 50%), allowing the financing party to use or sell the pledged collateral normally within the frozen percentage. Any amount exceeding this percentage will trigger an alarm and prevent the collateral from being released from the warehouse.

[0075] When the disposal action is to freeze all pledged assets, the system simultaneously sends a full freeze order to the supervisor's electronic access control system, electronic fence system, and alarm system, prohibiting any pledged assets from leaving the supervised area, and sends an emergency notification to the financing party and the funding party. For example, if a real risk event has a risk level of three and a risk confidence level of 0.6, and the disposal action is determined to be a partial freeze of pledged assets, the system sends an order to the warehouse access control system, allowing only 30% of the pledged assets to leave the warehouse. For the remaining 70% of the goods, the access control system will refuse entry and notify the supervisor.

[0076] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0077] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0078] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0079] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0080] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for monitoring movable property pledge based on the Internet of Things, characterized in that, Includes the following steps: By deploying IoT sensor nodes and environmental monitoring equipment on the pledged assets, multimodal perception data of each pledged entity is collected in real time, and the current value of each pledged entity and the trend of its change are dynamically calculated to generate a value vector containing the current value, value confidence, and value change trend. The value vectors are aggregated into a group value distribution, and a group behavior baseline is constructed based on the statistical characteristics of the group value distribution. The value vector of each pledged individual is compared with the group behavior baseline in real time to identify individuals that deviate from the group behavior baseline and mark the deviation as an anomaly to be confirmed. A list of anomalies to be confirmed containing the anomaly type, the degree of deviation, and the identifier of the associated individual is generated. For each unconfirmed abnormal event in the list of confirmed abnormal events, the source of the abnormal event is determined by the signal continuity in the multimodal sensing data, the sensor node status identifier, and the network communication log, whether it is a sensor node failure, communication interruption, or a change in the actual state of the pledged goods. Abnormal events originating from sensor node failure or communication interruption are marked as false alarm events and filtered out. Only abnormal events originating from changes in the actual state of the pledged assets are output as real risk events. Real risk events include risk level and risk confidence level. Based on the risk level and risk confidence of the actual risk event, the risk control strategy parameters for monitoring movable property pledge are dynamically adjusted. The risk control strategy parameters include the pledge ratio and the intensity of disposal actions. Based on the adjusted intensity of disposal actions, the corresponding disposal operations of the pledged property are executed.

2. The method for monitoring movable property pledge based on the Internet of Things according to claim 1, characterized in that, Multimodal sensing data includes at least physiological parameters and behavioral data, wherein the physiological parameters include body temperature and the behavioral data include weight, daily exercise or food intake.

3. The method for monitoring movable property pledge based on the Internet of Things according to claim 2, characterized in that, The intensity of the action is set in order of increasing risk level: only record reminder, reduce the pledge amount, partially freeze the pledged assets or freeze all the pledged assets, so as to achieve flexible adjustment that continuously matches the risk level.

4. The method for monitoring movable property pledge based on the Internet of Things according to claim 3, characterized in that, The value vector is generated as follows: Extract at least one feature data that can be used for valuation from the multimodal perception data of the pledged individual. The feature data includes weight, body temperature, daily exercise or food intake. Calculate the current individual value: The current individual value is equal to the individual's benchmark value multiplied by the product of the correction factors corresponding to each feature data. Each correction factor is linearly determined based on the degree of deviation between the corresponding feature data and the standard value. Specifically, the correction factor is equal to one minus the ratio of the absolute value of the deviation to the maximum allowable deviation. The value confidence level is calculated based on the signal continuity and data integrity rate of the sensor nodes: the value confidence level is equal to one minus the ratio of the number of missing data points to the theoretical number of data points, and then multiplied by the preset signal attenuation coefficient; the preset signal attenuation coefficient is obtained by continuously collecting data from a standard signal source in an interference-free environment and statistically analyzing the average signal attenuation per unit time. The value change trend is calculated by fitting a linear trend to multiple historical value values ​​within the most recent time window: the value change trend is equal to the difference between the current value and the initial value of the window divided by the window duration. The calculated current individual value, value confidence level, and value change trend are combined into a value vector.

5. The method for monitoring movable property pledge based on the Internet of Things according to claim 4, characterized in that, The process of constructing a baseline of group behavior based on the statistical characteristics of group value distribution is as follows: Obtain the value vectors of all pledged individuals within the same regulatory batch. Each value vector contains three dimensions: current individual value, value confidence level, and value change trend. Then, perform benchmark normalization on the current individual value of each individual, that is, divide the current individual value by the benchmark value of that individual to obtain the relative value. Calculate the mean and standard deviation of all individuals in the relative value dimension, the mean and standard deviation in the value confidence dimension, and the mean and standard deviation in the value change trend dimension; The average values ​​of the three dimensions are combined to form the center vector, and the standard deviations of the three dimensions are combined to form the range vector. The center vector and the range vector together constitute the baseline of group behavior.

6. The method for monitoring movable property pledge based on the Internet of Things according to claim 5, characterized in that, Identifying individuals who deviate from the group's behavioral baseline refers to: For each pledged individual, its relative value, value confidence, and value change trend are compared with the average value of the corresponding dimension in the group behavior baseline. The ratio of the deviation value of each dimension to the standard deviation of that dimension is calculated. The square root of the sum of the squares of the three ratios is used to obtain the comprehensive deviation distance. Determine whether the overall deviation distance is greater than a preset deviation threshold, which is set to two; if it is greater, mark the individual as an individual who deviates from the group behavior baseline, and the degree of deviation is the value of the overall deviation distance.

7. The method for monitoring movable property pledge based on the Internet of Things according to claim 6, characterized in that, The list of pending exception events is generated as follows: For individuals marked as deviating, the deviation multiples for each of the three dimensions—relative value, value confidence level, and value change trend—are calculated. The deviation multiple is the absolute value of the deviation in that dimension divided by the standard deviation of that dimension. The dimension with the largest deviation multiple among the three dimensions is selected as the primary deviation dimension; if multiple dimensions have the same largest deviation multiple, the primary deviation dimension is determined according to the priority order of value confidence, value change trend, and relative value. The type of abnormal event is determined based on the direction and attributes of the main deviation dimension: When the main deviation dimension is the value confidence level and its value is lower than the group average, the anomaly type is insufficient data credibility. When the main deviation dimension is the value confidence level and its value is higher than the group average, the anomaly event type is data abnormal stability. When the main deviation dimension is the value change trend and its value is lower than the group average, the abnormal event type is value growth stagnation; When the main deviation dimension is the value change trend and its value is higher than the group average, the abnormal event type is value growth that is too fast. When the main deviation dimension is relative value and its value is lower than the group average, the anomaly type is low value; When the main deviation dimension is a relative value and its value is higher than the group average, the anomaly type is value-biased. The abnormal event type, deviation degree, and individual identifier are combined into a single abnormal event to be confirmed. All abnormal events to be confirmed are then aggregated to generate a list of abnormal events to be confirmed.

8. The method for monitoring movable property pledge based on the Internet of Things according to claim 7, characterized in that, The internal process for distinguishing the source of an anomaly to be confirmed and outputting the actual risk event is as follows: For each unconfirmed abnormal event, acquire the multimodal sensing data of its associated individuals, and extract signal continuity parameters, sensor node status identifiers, and network communication logs from it; the signal continuity parameters include the ratio of the number of most recent consecutive valid data frames to the theoretical number of frames; Sensor node status indicators include node battery voltage and sensor data self-test status; network communication logs include base station signal strength and packet loss rate; The source of anomalies is distinguished according to the preset hierarchical judgment rules. Then, the abnormal events that are judged to be sensor node failures or communication interruptions are marked as false alarm events and deleted from the list of abnormal events to be confirmed. Only the abnormal events that are judged to be changes in the actual state of the pledged assets are retained as real risk events. Calculate the risk level and risk confidence level of a real risk event: The risk level is determined by rounding up the ratio of the comprehensive deviation distance to the deviation threshold; the risk confidence level is equal to one minus the false alarm rate, where the false alarm rate is the ratio of the number of filtered false alarm events to the total number of abnormal events to be confirmed in the regulatory batch.

9. A method for monitoring movable property pledge based on the Internet of Things according to claim 8, characterized in that, The preset level determination rules refer to: At the first level, if the battery voltage is lower than the operating voltage, the sensor node is considered to have failed. The second level is that if the battery voltage is not lower than the operating voltage but the signal continuity parameter is lower than the first threshold, then when the packet loss rate is higher than the second threshold or the base station signal strength is lower than the third threshold, it is determined that the communication is interrupted; otherwise, it is determined that the sensor node is faulty. At the third level, if the battery voltage is not lower than the operating voltage and the signal continuity parameter is not lower than the first threshold, but the sensor data self-test status shows that the measured value exceeds the calibration range, then the sensor node is determined to be faulty. At the fourth level, if none of the above conditions are met, it is determined that the actual state of the pledged property has changed.

10. A method for monitoring movable property pledge based on the Internet of Things according to claim 9, characterized in that, Dynamically adjust the risk control strategy parameters for monitoring movable asset pledges: Obtain the risk level and risk confidence level of a real risk event, where the risk level is an integer greater than or equal to one, and the risk confidence level is a real number greater than or equal to zero and less than or equal to one; The collateral ratio is adjusted based on the risk level and risk confidence level: the new collateral ratio is equal to the original collateral ratio multiplied by the adjustment factor, which is one minus the ratio of the current risk level to the maximum risk level, and then multiplied by the risk confidence level, where the maximum risk level is set to four. The intensity of the response is determined based on the risk level: when the risk level is one, the intensity of the response is to only record and remind the user. When the risk level is two, the intensity of the action is to reduce the amount of pledged collateral. When the risk level is three, the intensity of the action is to partially freeze the pledged assets; When the risk level is greater than or equal to four, the intensity of the action is to freeze all pledged assets; at the same time, the intensity of the action is adjusted according to the risk confidence level: if the risk confidence level is less than 0.5, the intensity of the action is reduced by one level; if the intensity of the action before the reduction is already only to record a reminder, then no reduction is made.

Images