Multi-mode switching low-power-consumption Internet of Things equipment management system

Abstract

The invention relates to the field of Internet of Things equipment power consumption management, in particular to a multi-mode switching low-power-consumption Internet of Things equipment management system. By setting a multi-layer electronic fence area and combining behavioral habit modeling and border crossing intention prediction, intelligent judgment and switching of equipment operation modes are realized: when equipment is in a first electronic fence, a low-power-consumption mode is entered; when entering a buffer area between the first electronic fence and the second electronic fence, predicting whether a border crossing intention exists or not based on the multi-dimensional behavior characteristics, and judging whether to switch to an activity tracking mode or not; and when the second electronic fence is exceeded, determining whether to enter a warning mode or not by combining the border crossing behavior intensity and the historical precedent assessment risk. The system also supports adaptive adjustment of sampling frequency and dynamic switching of communication protocols. According to the invention, the problems of frequent switching and high energy consumption caused by boundary oscillation in the prior art are effectively solved, and low-power-consumption, intelligent and high-reliability operation of the positioning equipment in nursing application is realized.

Description

Technical Field

[0001] The present invention relates to the field of power consumption management of Internet of Things devices, and in particular to a multi-mode switching low-power Internet of Things device management system. Background Art

[0002] With the development of IoT technology, tracking devices are widely used in scenarios such as personal care, pet management, and asset tracking. Wearable positioning terminals based on GNSS and cellular communication networks have become mainstream products. These devices often incorporate electronic fencing functionality, sounding an alarm when a monitored object (such as a pet) leaves a designated safe zone, prompting the user to take appropriate action.

[0003] However, existing technologies have significant flaws. On the one hand, traditional trackers often use a fixed-frequency data collection and upload strategy, lacking intelligent perception of object behavior patterns. This makes it easy to upload data frequently when it is not necessary, resulting in a waste of resources. On the other hand, existing systems often immediately switch the device's operating mode and trigger an alarm when entering or leaving the boundary of an electronic fence, without making a reasonable judgment on the carrier's actual intentions. Especially in cases where the monitored object wanders back and forth within the boundary area of ​​the electronic fence, such as a pet moving around the edge of the yard, existing technologies are prone to frequently triggering electronic fence events, causing the device to constantly switch between low-power and high-frequency sampling modes. This "boundary oscillation" phenomenon not only causes invalid alarms, but also leads to frequent device wake-ups and high-frequency data uploads, which significantly increases power consumption, shortens battery life, and seriously affects system stability and user experience.

[0004] Therefore, there is an urgent need for an IoT device management system with multi-mode intelligent switching capabilities, which can intelligently judge the working status based on the object behavior pattern and out-of-bounds intention, and avoid unnecessary energy consumption and false alarms caused by frequent changes in location boundaries. Summary of the Invention

[0005] The present invention provides a multi-mode switching low-power consumption Internet of Things device management system.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A multi-mode switching low-power IoT device management system, comprising:

[0008] A data acquisition module, used to collect acceleration information, location information and timestamps of IoT devices at a preset frequency;

[0009] A low-power mode determination module, configured to perform a static determination based on the positional relationship between the device's current location and the first electronic fence, and maintain low-power mode and perform data sampling at a preset frequency when the device is within the first electronic fence;

[0010] An activity tracking mode determination module is configured to construct a behavior habit model based on data sampled at a preset frequency when the device's location exceeds the first electronic fence and enters the buffer zone between the first and second electronic fences. The behavior habit model predicts whether the carrier has an intention to cross the boundary. If the prediction result indicates an intention to cross the boundary, the device switches to activity tracking mode and adaptively increases the sampling frequency; otherwise, the device maintains a low power consumption mode.

[0011] The alert mode identification module is used to determine the risk when the carrier exceeds the second electronic fence, combining the intensity score of the crossing behavior and the probability of historical crossing precedents. If it is determined to be a high-risk state, the device switches to alert mode, starts continuous positioning recording and pushes alarm information; otherwise, it maintains activity tracking mode.

[0012] Furthermore, the steps to construct the behavior habit model are:

[0013] Clean and align the acceleration information, position information and timestamp collected at a preset frequency in the buffer;

[0014] Extracting behavioral samples based on the carrier's multidimensional features at multiple different time periods, the multidimensional features including path trajectory, hot spots, activity frequency, and acceleration fluctuation amplitude;

[0015] Perform cluster analysis on the extracted behavioral samples, identify high-frequency behavioral patterns and their distribution, and construct a probabilistic graphical model to describe behavioral state transitions;

[0016] The newly collected real-time behavior data is input into the probability graph model, and a probability score of the boundary crossing intention is output.

[0017] Furthermore, the probabilistic graphical model is a hidden Markov model, and the steps of outputting the probability score of the crossing intention are:

[0018] Modeling the high-frequency behavioral patterns and their distributions identified by cluster analysis as a set of latent states, including stillness, regular activity, exploratory behavior, and boundary wandering;

[0019] The acceleration change, path change rate and position offset are input as observation sequences;

[0020] Train the hidden state transition probability matrix and observation probability matrix based on historical behavior samples;

[0021] The Viterbi algorithm is used to infer the real-time behavior observation sequence and calculate the probability that the carrier is in the cross-boundary related hidden state;

[0022] If the probability of crossing the boundary intention exceeds the crossing the boundary intention threshold, it is determined that there is a tendency to cross the boundary, triggering the device mode switch.

[0023] Furthermore, the steps of adaptively increasing the sampling frequency are:

[0024] Calculating the distance between the device and the second electronic fence based on the relative position relationship between the current position of the device and the first electronic fence and the second electronic fence;

[0025] When it is predicted that the carrier intends to cross the boundary and the device is continuously approaching the second electronic fence, the data sampling frequency is increased in a linear function manner according to the distance value;

[0026] When the device location is far away from the second electronic fence, or the probability score of the crossing intention drops below the preset threshold, the preset sampling frequency is restored.

[0027] Furthermore, the intensity score of the out-of-bounds behavior is calculated based on the carrier's continuous movement direction, average speed and projected distance in the direction of the second electronic fence in the buffer zone; the probability of historical out-of-bounds precedents is calculated by analyzing the proportion of the carrier's historical out-of-bounds behaviors in the same time period.

[0028] Furthermore, the risk determination step includes:

[0029] Assign preset weights to the intensity score of the transgression behavior and the probability of historical transgression precedents, and calculate the risk assessment coefficient;

[0030] If the risk assessment coefficient is greater than the risk threshold, it is judged to be a high-risk state; otherwise, it is judged to be a low-risk state.

[0031] Furthermore, in low-power mode, the IoT device uses the BLE protocol for short-range, low-frequency communication; in activity tracking mode, the IoT device switches to the NB-IoT protocol for medium-frequency, low-latency data upload; in alert mode, the IoT device switches to the collaborative operation of NB-IoT and cellular networks to ensure high-reliability, low-latency data transmission and remote alarm push.

[0032] Furthermore, the system also includes an electronic fence setting module for initializing and maintaining the first and second electronic fences.

[0033] The beneficial effects of the present invention are:

[0034] 1. This invention establishes multiple layers of electronic fences and incorporates behavioral habit models and boundary-crossing intention prediction mechanisms, enabling devices to intelligently switch between low-power, activity tracking, and alert modes based on actual conditions. This prevents frequent mode switching and invalid alarms caused by proximity to boundaries. This strategy significantly reduces device power consumption in normal states, effectively extending battery life and making it particularly suitable for large-scale deployments of devices in pet care scenarios.

[0035] 2. This invention introduces a method for determining border crossing risk based on multi-dimensional behavioral characteristics, comprehensively considering movement direction, speed characteristics, and the probability of historical border crossings to accurately identify the carrier's intention to cross the border. Compared to traditional location-based mode switching mechanisms, this system significantly improves behavior prediction accuracy and response sensitivity, avoiding false positives and missed negatives, and improving device reliability and user experience in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0037] Figure 1 This is a structural diagram of a multi-mode switching low-power Internet of Things device management system provided by the present invention;

[0038] Figure 2 This is an execution flow chart of a multi-mode switching low-power Internet of Things device management system provided by the present invention. DETAILED DESCRIPTION

[0039] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0040] Example 1

[0041] A multi-mode switching low-power IoT device management system, such as Figure 1 As shown, including:

[0042] A data acquisition module, used to collect acceleration information, location information and timestamps of IoT devices at a preset frequency;

[0043] Specifically, a three-axis MEMS accelerometer is used to collect acceleration information in the X, Y, and Z directions; a multi-mode GNSS receiver chip is used to obtain location information such as latitude and longitude, altitude, and speed; and the standard time provided by the GPS module is used as the sampling timestamp.

[0044] A low-power mode determination module, configured to perform a static determination based on the positional relationship between the device's current location and the first electronic fence, and maintain low-power mode and perform data sampling at a preset frequency when the device is within the first electronic fence;

[0045] Specifically, the first electronic fence is set as a circular area, consisting of the latitude and longitude coordinates of the center point and a radius. After the device obtains the current location coordinates through the GNSS module, it uses a spherical distance calculation method to determine the distance between the current location and the center point of the fence. If this distance is less than or equal to the preset radius, the device is determined to be within the electronic fence and enters or maintains low power mode.

[0046] An activity tracking mode determination module is configured to construct a behavior habit model based on data sampled at a preset frequency when the device's location exceeds the first electronic fence and enters the buffer zone between the first and second electronic fences. The behavior habit model predicts whether the carrier has an intention to cross the boundary. If the prediction result indicates an intention to cross the boundary, the device switches to activity tracking mode and adaptively increases the sampling frequency; otherwise, the device maintains a low power consumption mode.

[0047] Furthermore, the steps to construct the behavior habit model are:

[0048] Clean and align the acceleration information, position information and timestamp collected at a preset frequency in the buffer;

[0049] Extracting behavioral samples based on the carrier's multidimensional features at multiple different time periods, the multidimensional features including path trajectory, hot spots, activity frequency, and acceleration fluctuation amplitude;

[0050] Perform cluster analysis on the extracted behavioral samples, identify high-frequency behavioral patterns and their distribution, and construct a probabilistic graphical model to describe behavioral state transitions;

[0051] The newly collected real-time behavior data is input into the probability graph model, and a probability score of the boundary crossing intention is output.

[0052] Specifically, when a device enters the buffer zone between the first and second electronic fences, the system collects the carrier's acceleration information, location information, and corresponding timestamps at a preset frequency. The collected data is first cleaned to remove outliers and invalid segments, and data from different sources is aligned in a unified time series to ensure the accuracy and consistency of subsequent analysis. Based on the preprocessed data, the system extracts multidimensional behavioral information reflecting the carrier's activity characteristics, including path trajectories, hot spots, activity frequency, and acceleration fluctuation amplitude, as basic features for behavioral modeling. The path trajectory is constructed by connecting consecutive location information in chronological order to form movement trajectory segments, and further calculating parameters such as directional change rate, curvature, and path length. The hotspot detection method uses density-based clustering algorithms (such as DBSCAN) on location information to identify spatial clusters where carriers spend extended periods of time, and then calculates the visit frequency and duration of each hotspot. Activity frequency assesses daily activity by measuring the number of location points or significant acceleration fluctuations per unit time. Acceleration fluctuation amplitude is determined by calculating metrics such as the mean, standard deviation, and coefficient of variation of the three-axis acceleration modulus to determine whether the carrier is stationary, walking, or moving rapidly. Based on these extracted features, the system constructs behavioral samples from data across multiple time periods and analyzes these samples using a clustering algorithm to identify high-frequency behavioral patterns of carriers. These patterns reflect their daily spatial activity patterns and are used to construct a profile of their behavioral preferences. After identifying common behavioral patterns, the system further constructs a hidden Markov probabilistic graphical model describing the state transition relationships between these patterns, establishing dynamic transition patterns between behavioral sequences. After the newly collected real-time behavior data is input into the model, a probability score of the cross-border intention can be output for use in subsequent mode switching decisions.

[0053] By building a behavioral habit model based on multi-dimensional sensor data collected within the buffer zone, we can deeply explore the carrier's activity patterns and trajectory characteristics over different time periods. Cluster analysis identifies high-frequency behavioral patterns, and a probabilistic graphical model models the transition trends of behavioral states, enabling real-time prediction and probabilistic quantification of out-of-bounds intentions. This approach not only enhances the system's accuracy in detecting abnormal behavior but also significantly improves the intelligence and forward-looking judgment capabilities of mode switching, effectively balancing the needs of power consumption control and real-time behavior monitoring.

[0054] Furthermore, the probabilistic graphical model is a hidden Markov model, and the steps of outputting the probability score of the crossing intention are:

[0055] Modeling the high-frequency behavioral patterns and their distributions identified by cluster analysis as a set of latent states, including stillness, regular activity, exploratory behavior, and boundary wandering;

[0056] The acceleration change, path change rate and position offset are input as observation sequences;

[0057] Train the hidden state transition probability matrix and observation probability matrix based on historical behavior samples;

[0058] The Viterbi algorithm is used to infer the real-time behavior observation sequence and calculate the probability that the carrier is in the cross-boundary related hidden state;

[0059] If the probability of crossing the boundary intention exceeds the crossing the boundary intention threshold, it is determined that there is a tendency to cross the boundary, triggering the device mode switch.

[0060] Specifically, the system abstractly models the carriers' frequent behavioral patterns identified through cluster analysis into a set of hidden states through semantic labeling and state encoding mapping. This set includes four main behavioral states: stationary state (e.g., prolonged dwelling), regular activity state (e.g., entering and exiting a predetermined route), exploratory behavior state (e.g., occasionally leaving the usual area), and border wandering state (e.g., frequently approaching the edge of the second electronic fence). The system then inputs the collected acceleration changes, path change rates, and position offsets as observation sequences into a hidden Markov model. Acceleration changes are represented by the rate of change of the acceleration modulus; path change rates are calculated by the frequency of trajectory direction changes; and position offsets represent the offset distance of the current position relative to frequent dwell points or the center of the first fence. Next, based on historical behavioral samples, the system uses the Baum-Welch algorithm to train the parameters of the hidden Markov model, including the state transition probability matrix and the observation probability matrix. This training process constructs transition patterns between behavioral states and the conditional probability distribution from a specific state to a specific observation value. During operation, the system uses the Viterbi algorithm to infer the most likely state path from the real-time behavioral observation sequence. It then combines the probability output to calculate the probability that the user is currently in a boundary-crossing state, such as "exploratory behavior" or "boundary wandering." If this probability exceeds a preset boundary-crossing intent threshold (e.g., 0.75), the system determines that the user has a tendency to cross the boundary and immediately triggers the device to switch from low-power mode to activity tracking mode to improve data sampling frequency and responsiveness.

[0061] By introducing a hidden Markov model to model the carrier's behavior, the system abstracts complex and ever-changing behavior patterns into a finite set of hidden states. Combined with multi-dimensional observational information such as acceleration changes, path change rates, and position offsets, the system dynamically tracks and probabilistically predicts state transitions. Leveraging the Viterbi algorithm to infer hidden states from real-time behavioral data, the system accurately identifies whether the carrier is in a behavior phase with potential cross-border intent, thereby triggering device mode switching in advance and significantly improving the sensitivity and timeliness of cross-border behavior detection.

[0062] Furthermore, the steps of adaptively increasing the sampling frequency are:

[0063] Calculating the distance between the device and the second electronic fence based on the relative position relationship between the current position of the device and the first electronic fence and the second electronic fence;

[0064] When it is predicted that the carrier intends to cross the boundary and the device is continuously approaching the second electronic fence, the data sampling frequency is increased in a linear function manner according to the distance value;

[0065] When the device location is far away from the second electronic fence, or the probability score of the crossing intention drops below the preset threshold, the preset sampling frequency is restored.

[0066] Specifically, the system calculates the Euclidean distance between the device's current location and the boundary of the second electronic fence based on the current GNSS positioning information, and combines this with a determination of whether the device is continuously approaching the second electronic fence. This determination can be made based on whether the position change vector's direction points toward the fence boundary within five or more consecutive time windows. To increase the sampling frequency and more intensively monitor suspicious activity, the system employs the following linear function to increase the sampling frequency:

[0067] ;

[0068] in, Indicates the Euclidean distance between the current position and the boundary of the second geo-fence. Indicates the currently calculated sampling frequency, Indicates the default sampling frequency in low power mode. Indicates the highest sampling frequency allowed by the activity tracking mode, which is set to , Indicates the radius of the buffer zone, that is, the upper limit of the distance between the first fence and the second fence. Additionally, the system terminates the frequency increase and restores the sampling frequency to the default value if any of the following conditions are met: the real-time predicted probability score of boundary crossing intention drops below the preset threshold, or the device moves away from the second fence for five or more consecutive time windows.

[0069] By introducing an adaptive sampling frequency adjustment mechanism based on both distance and intent, the system dynamically increases the data sampling frequency according to a linear function when the carrier approaches the border and intends to cross, thereby more precisely capturing behavioral changes. When the risk decreases or the distance increases, it automatically reverts to a low-power sampling mode. This approach balances the response accuracy of behavioral monitoring with device energy consumption control, improving the system's intelligent adjustment capabilities and operational efficiency in complex boundary environments.

[0070] The alert mode identification module is used to determine the risk when the carrier exceeds the second electronic fence, combining the intensity score of the crossing behavior and the probability of historical crossing precedents. If it is determined to be a high-risk state, the device switches to alert mode, starts continuous positioning recording and pushes alarm information; otherwise, it maintains activity tracking mode.

[0071] Furthermore, the intensity score of the out-of-bounds behavior is calculated based on the carrier's continuous movement direction, average speed and projected distance in the direction of the second electronic fence in the buffer zone; the probability of historical out-of-bounds precedents is calculated by analyzing the proportion of the carrier's historical out-of-bounds behaviors in the same time period.

[0072] By combining the carrier's movement direction, average speed and projected distance toward the boundary within the buffer zone to calculate the intensity score of the cross-border behavior, and introducing the frequency of historical cross-border behaviors in the same period as the precedent probability, the system effectively improves its comprehensive judgment ability on potential cross-border risks.

[0073] Specifically, the system first analyzes the data from the time the carrier enters the buffer zone to the time the border crossing occurs, and extracts three key features: continuous motion direction stability, normalizing the displacement direction vector in each time window, and calculating the mean cosine similarity in multiple continuous time windows to reflect the consistency of the motion direction; speed mean, calculating the displacement distance per unit time based on continuous position information, taking the sliding window average, and evaluating the intensity of the movement; projection distance in the direction of the fence, calculating the unit normal projection distance of each position point relative to the fence boundary, and finding the average value to indicate whether the direction of movement is clearly towards the boundary. After the three features are normalized, a weighted linear combination is used to form a score for the intensity of the border crossing behavior. ,in represents the normalized directional consistency score, represents the normalized mean velocity, The system retrieves the carrier's historical behavior records and counts whether the carrier has crossed the border in the current time period (such as the same hour of each day). If a certain period is monitored N times, and the crossing of the border occurs n times, the probability of the historical crossing of the border precedent is expressed as .

[0074] Furthermore, the risk determination step includes:

[0075] Assign preset weights to the intensity score of the transgression behavior and the probability of historical transgression precedents, and calculate the risk assessment coefficient;

[0076] If the risk assessment coefficient is greater than the risk threshold, it is judged to be a high-risk state; otherwise, it is judged to be a low-risk state.

[0077] Specifically, the system combines the two indicators of the intensity score of the cross-border behavior and the probability of historical cross-border precedents to perform risk scoring and obtain the risk assessment coefficient. ,in, is the weight coefficient, The more times a certain area has been breached, the lower the risk. If the risk assessment coefficient is greater than the risk threshold (e.g., 0.75), the area is considered high risk; otherwise, it is considered low risk.

[0078] By assigning preset weights to the intensity of boundary violations and the probability of historical boundary violations, and calculating a risk assessment coefficient, the system implements a multi-factor quantitative risk assessment mechanism, effectively avoiding misjudgments caused by a single indicator. This approach enhances the flexibility and adjustability of risk assessment, enabling the device to more accurately identify high-risk boundary violations and promptly switch to alert mode in real-world applications.

[0079] Furthermore, in low-power mode, the IoT device uses the BLE protocol for short-range, low-frequency communication; in activity tracking mode, the IoT device switches to the NB-IoT protocol for medium-frequency, low-latency data upload; in alert mode, the IoT device switches to the collaborative operation of NB-IoT and cellular networks to ensure high-reliability, low-latency data transmission and remote alarm push.

[0080] By intelligently switching communication protocols based on different operating modes, the system uses BLE for low-frequency, ultra-low-energy, short-range communication in low-power mode, NB-IoT for medium-frequency, low-latency data uploads in activity tracking mode, and a combination of NB-IoT and cellular networks in alert mode to ensure high reliability and long-range, real-time transmission of critical data. This layered communication strategy effectively balances power consumption, transmission efficiency, and timely alerts.

[0081] Furthermore, the system also includes an electronic fence setting module for initializing and maintaining the first and second electronic fences.

[0082] By introducing the electronic fence setting module, the system can flexibly initialize and maintain the first and second electronic fences, realizing dynamic control and behavioral boundary definition of different geographical areas.

[0083] Example 2

[0084] This embodiment provides a multi-mode switching low-power IoT device management system for pet smart locators. The system performs the following process: Figure 2 The system is deployed in a wearable positioning device worn around the pet's neck, helping users to understand the pet's activity status in real time, saving energy, and providing timely alarms when abnormal behavior occurs.

[0085] The system operation process is as follows:

[0086] The device's built-in triaxial MEMS accelerometer and multi-mode GNSS receiver chip collect the pet's current triaxial acceleration, latitude, longitude, altitude, and device timestamp at a preset frequency of once per minute. GPS time synchronization ensures the timing alignment of different data sources.

[0087] When the system determines that the pet is in the "home" area set by the user (i.e. the first electronic fence), such as the owner's residence or yard (set by the center point and radius of the circular electronic fence), the device enters low-power mode, only collecting data at a low frequency, and using the BLE protocol for short-range communication with the owner's mobile phone to extend battery life.

[0088] When a pet leaves the "home" fence and enters the "buffer zone" (the area between the first and second electronic fences, for example, from "home" to the "cell edge"), the system initiates behavioral modeling. The device continues to collect data at a moderate frequency and extracts features such as whether the pet is moving toward the fence boundary or wandering around.

[0089] A hidden Markov model generated through historical modeling determines whether the user is "expected to leave" or "intended to cross the boundary" (e.g., frequent proximity to the fence edge, abnormal acceleration). If the intention score exceeds the threshold, the device automatically switches to activity tracking mode and adaptively increases the sampling frequency based on the distance from the second fence.

[0090] When a pet crosses a secondary electronic fence (e.g., at the edge of a residential area), the system calculates a risk assessment coefficient based on its average speed and directional trends within the buffer zone, as well as whether similar crossings have occurred repeatedly during the same time period. If this coefficient exceeds a threshold (e.g., 0.75), the system enters alert mode, enabling continuous GNSS positioning and uploading location information to the cloud every 10 seconds. Through NB-IoT and cellular network communication, the system sends real-time crossing alerts to the owner's mobile app or monitoring platform.

[0091] The system in this embodiment dynamically adjusts its operating mode based on pet behavior and location, achieving energy-saving operation, intelligent perception, and efficient response. This not only improves device battery life but also reduces false alarm rates, enabling immediate response when a pet actually crosses a boundary or becomes lost. It is widely applicable to scenarios such as home pet care and free-range pastures.

[0092] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A multi-mode switching low-power Internet of Things device management system, characterized in that: include: A data acquisition module, used to collect acceleration information, location information and timestamps of IoT devices at a preset frequency; A low-power mode determination module, configured to perform a static determination based on the positional relationship between the device's current location and the first electronic fence, and maintain low-power mode and perform data sampling at a preset frequency when the device is within the first electronic fence; An activity tracking mode discrimination module is configured to construct a behavior habit model based on data sampled at a preset frequency when the device's location exceeds the first electronic fence and enters the buffer zone between the first and second electronic fences; the behavior habit model predicts whether the carrier has the intention to cross the boundary; If the prediction result indicates that there is an out-of-bounds intention, the device switches to activity tracking mode and adaptively increases the sampling frequency; otherwise, it maintains low power consumption mode; The alert mode identification module is used to determine the risk when the carrier exceeds the second electronic fence by combining the intensity score of the crossing behavior and the probability of historical crossing precedents; If it is determined to be a high-risk state, the device switches to alert mode, starts continuous positioning recording and pushes alarm information; otherwise, it maintains activity tracking mode.

2. A multi-mode switching low-power Internet of Things device management system according to claim 1, characterized in that: The steps to build a behavioral habit model are: Clean and align the acceleration information, position information and timestamp collected at a preset frequency in the buffer; Extracting behavioral samples based on the carrier's multidimensional features at multiple different time periods, the multidimensional features including path trajectory, hot spots, activity frequency, and acceleration fluctuation amplitude; Perform cluster analysis on the extracted behavioral samples, identify high-frequency behavioral patterns and their distribution, and construct a probabilistic graphical model to describe behavioral state transitions; The newly collected real-time behavior data is input into the probability graph model, and a probability score of the boundary crossing intention is output.

3. A multi-mode switching low-power Internet of Things device management system according to claim 2, characterized in that: The probabilistic graphical model is a hidden Markov model, and the steps for outputting the probability score of the crossing intention are as follows: Modeling the high-frequency behavioral patterns and their distributions identified by cluster analysis as a set of latent states, including stillness, regular activity, exploratory behavior, and boundary wandering; The acceleration change, path change rate and position offset are input as observation sequences; Train the hidden state transition probability matrix and observation probability matrix based on historical behavior samples; The Viterbi algorithm is used to infer the real-time behavior observation sequence and calculate the probability that the carrier is in the cross-boundary related hidden state; If the probability of crossing the boundary intention exceeds the crossing the boundary intention threshold, it is determined that there is a tendency to cross the boundary, triggering the device mode switch.

4. The multi-mode switching low-power Internet of Things device management system according to claim 1, characterized in that: The steps to adaptively increase the sampling frequency are: Calculating the distance between the device and the second electronic fence based on the relative position relationship between the current position of the device and the first electronic fence and the second electronic fence; When it is predicted that the carrier intends to cross the boundary and the device is continuously approaching the second electronic fence, the data sampling frequency is increased in a linear function manner according to the distance value; When the device location is far away from the second electronic fence, or the probability score of the crossing intention drops below the preset threshold, the preset sampling frequency is restored.

5. The multi-mode switching low-power Internet of Things device management system according to claim 1, characterized in that: The intensity score of the out-of-bounds behavior is calculated based on the carrier's continuous movement direction, average speed and projected distance to the second electronic fence in the buffer zone; the probability of historical out-of-bounds precedents is calculated by analyzing the proportion of the carrier's historical out-of-bounds behaviors in the same time period.

6. The multi-mode switching low-power Internet of Things device management system according to claim 1, characterized in that: The steps of risk determination include: Assign preset weights to the intensity score of the transgression behavior and the probability of historical transgression precedents, and calculate the risk assessment coefficient; If the risk assessment coefficient is greater than the risk threshold, it is judged to be a high-risk state; otherwise, it is judged to be a low-risk state.

7. The multi-mode switching low-power Internet of Things device management system according to claim 1, characterized in that: In low-power mode, the IoT device uses the BLE protocol for short-range, low-frequency communication; In activity tracking mode, the IoT device switches to the NB-IoT protocol for medium-frequency, low-latency data upload; In alert mode, the IoT device switches to collaborative operation of NB-IoT and cellular networks to ensure high reliability, low latency data transmission and remote alarm push.

8. The multi-mode switching low-power consumption Internet of Things device management system according to claim 1, characterized in that: The system further comprises an electronic fence setting module for initializing and maintaining the first and second electronic fences.

Images