Intelligent power adapter based on Internet of Things and working method thereof

Abstract

The invention discloses an intelligent power adapter based on the Internet of Things and a working method thereof, a complete closed-loop system from risk perception, active protection to timely alarm is constructed through deep fusion of the Internet of Things and intelligent sensing, and the intelligent power adapter is switched to a safe power supply mode immediately, so that the intelligent power adapter is safe and reliable. The electric shock injury possibly suffered by the pet is fundamentally avoided, and the core potential safety hazard is solved; meanwhile, the event details are pushed to the user terminal in real time through the Internet of Things module, it is ensured that the user can know the situation at the first time and conduct remote processing, and the problem of reminding lag is effectively solved; in addition, the method has self-learning and self-adaptive optimization capabilities, and can continuously accumulate data to improve the recognition accuracy and adapt to user habits; finally, the whole scheme actively and intelligently guarantees pet safety, equipment safety and electricity utilization continuity without physical intervention of a user, and the use experience and psychological safety of a pet raising family are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power adapter technology, specifically to an Internet of Things-based smart power adapter and its operating method. Background Technology

[0002] Power adapters, as key components that convert AC mains power into DC power required by electronic devices, are widely used in various household appliances, digital products, and smart devices, becoming an indispensable power interface device in modern life. A power adapter typically consists of a housing containing rectification, filtering, transformation, and voltage regulation circuits, as well as an input power cord and an output DC cable. Its core design goal is to achieve efficient and stable power conversion and supply. However, with the deep integration of IoT technology and smart home scenarios, the functional positioning of power adapters is no longer limited to a simple "energy converter." Users' demands for its safety, interactivity, and intelligence are increasing.

[0003] Especially in pet-owning households, traditional power adapters expose a long-standing but unresolved safety hazard: exposed output cables, due to their shape, material, or the natural chewing instincts of pets, are easily targeted by pets (such as cats and dogs) for chewing and scratching. During this process, sharp teeth can pierce the cable insulation, causing direct contact with the internal live conductors, posing a very high risk of electric shock and seriously threatening the pet's life. Simultaneously, cable damage can also lead to electrical short circuits, equipment malfunctions, or even fires, causing property damage. Currently, common solutions to this problem are mostly passive protection measures, such as using cable sleeves or placing the adapter at a high place out of reach of pets. These methods not only rely on the user's constant vigilance, offering limited protection and a poor user experience, but also fail to take any proactive protective measures once the pet has started chewing, and cannot promptly notify users who may not be present when an incident occurs.

[0004] In recent years, while some technologies have attempted to add temperature monitoring or simple current detection to adapters to prevent overheating or short circuits, these solutions are "remedial measures," meaning they only act when an electrical fault has occurred or is about to occur. They cannot intervene during the physical damage stage when a pet's teeth come into contact with the cable, and therefore cannot fundamentally prevent pets from getting electric shocks. On the other hand, although IoT technology has been introduced into the home appliance field to enable remote status monitoring, no solution has yet been found that can systematically integrate it with the deep safety protection mechanism of the adapter to achieve a complete closed loop from "precise perception of risky behavior" to "instantaneous active power-off protection" and then to "real-time remote synchronous alarm."

[0005] Therefore, there is a significant gap in the relevant technology: there is a lack of a power adapter operating method that can intelligently identify the specific risky behavior of pet biting in real time, activate active electrical isolation instantly before the injury occurs, and simultaneously notify the user in real time via the Internet of Things and perform intelligent learning and optimization. Summary of the Invention

[0006] To overcome the shortcomings mentioned above, this invention aims to provide a comprehensive solution that integrates intelligent sensing, edge computing, active safety, and IoT communication, so as to fundamentally improve electrical safety and user experience in pet-owning households.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for operating an IoT-based smart power adapter includes the following steps: S100: By using a sensor array installed on the output line of the power adapter, the physical contact signal on the output line is monitored in real time, and the physical contact signal is sent to the control unit; The sensor array includes a flexible pressure sensor, a distributed strain sensor, and an impedance measurement sensor, which are embedded in the outer sheath of the output line in a matrix configuration. S200: The control unit analyzes the physical contact signal based on a pattern recognition algorithm, extracts contact features, and compares them with a preset pet biting pattern library to identify whether it is a pet biting behavior; S300: When a pet biting behavior is detected, the control unit controls the power output unit to switch to safe mode within milliseconds; S400: The IoT module encapsulates pet biting incidents and related information into alarm messages and wirelessly sends them to the user terminal to provide user alerts; The relevant information includes the time of the event, the location of the power adapter, and the location of the bite. S500: In safe mode, a self-test program is initiated to assess the degree of damage to the output line, and the pattern recognition algorithm is updated based on historical bite data to adaptively optimize the bite recognition accuracy. S600: Based on instructions from the user terminal, adjust the operating status of the power adapter, including restoring normal power output or maintaining a safe mode and triggering a local alarm.

[0008] As a further aspect of the present invention, step S100 specifically includes the following steps: S101: The output line is divided into multiple independent sensing segments along its length direction, and the sensor array includes a set of sensor units disposed in each sensing segment; Each sensor unit includes at least one of the flexible pressure sensors, one of the distributed strain sensors, one impedance measurement sensor, and one microcontroller node; S102: Each of the sensor units synchronously acquires the real-time physical contact signal of its sensing segment at a preset sampling frequency, wherein the real-time physical contact signal includes a pressure distribution map, strain rate of change and raw waveform data; S103: Each microcontroller node performs local preprocessing on the acquired real-time physical contact signal, extracts time-domain and frequency-domain features, and sends the preprocessed data to the control unit in real time.

[0009] As a further aspect of the present invention, step S200 specifically includes the following steps: S201: The control unit receives preprocessed data from each sensing segment and performs data fusion to generate a time-stamped matrix of the full contact status of the output line. S202: Extract at least two dynamic identification features from the full-line contact state matrix, including: tearing waveform features: calculate the impact steepness and period of the waveform based on the temporal correlation between the strain rate of change and the pressure peak. Movement pattern characteristics: Based on the propagation direction and speed of the pressure distribution map between adjacent sensing segments, identify whether it conforms to the movement pattern of tooth biting; Energy concentration characteristics: Analyze whether the energy of the sensor signal is concentrated in one or more local sensing segments within a preset time window; S203: Perform multi-level matching and judgment between the extracted dynamic recognition features and the feature vectors in the preset pet biting pattern library; First, a preliminary screening is performed to determine whether the impact steepness exceeds a first threshold and whether the energy concentration is higher than a second threshold. If the initial screening is passed, a secondary matching is performed, using a pattern recognition algorithm to calculate the similarity between the current dynamic recognition features and typical biting patterns, playful scratching patterns, and human plugging / unplugging patterns in the pet biting pattern library. S204: Based on the results of the multi-level matching judgment, if the similarity of the biting pattern output by the pattern recognition algorithm exceeds the set confidence level, and at the same time the movement pattern feature conforms to the preset biting path model, then it is finally determined to be a pet biting behavior, and a judgment result containing the biting intensity and the suspected pet body type is generated.

[0010] As a further aspect of the present invention, step S300 specifically includes the following steps: S301: After determining that a pet biting behavior has occurred, the control unit sends an emergency switching command containing biting location information to the power management unit and the power output unit. S302: The power management unit performs a millisecond-level safe mode switching operation according to the emergency switching command: Immediately disconnect the standard operating voltage and current supplied by the power output unit to the output line; S303: The control unit, in conjunction with the power management unit, applies a low voltage to the current output path of the power adapter and simultaneously switches the entire power adapter to a low-power operation state (based on the current production quality standards of the output line, it is unlikely that a common household pet will bite through the output line in one go, so the current output path of the power adapter remains effective. For a common household pet, it would take a certain amount of time to completely bite through the copper and aluminum wire cores). S304: After the safety mode is activated, based on the sensor array, the cable damage diagnosis of the output line is initiated: by injecting low-voltage test signals into both ends of the output line and combining the impedance change data fed back by the sensor array, the degree of cable damage and insulation status centered on the biting position are analyzed. S305: Based on the cable damage level, historical tearing event data, and current IoT connection status, execute adaptive security policy adjustments: If the diagnostic results indicate that the insulation layer is damaged, the power adapter will be locked in safe mode and automatic recovery will be disabled. If the damage is minor and the network connection is stable, the priority of the IoT alarm should be increased while maintaining low voltage output.

[0011] As a further aspect of the present invention, step S400 specifically includes the following steps: S401: The control unit encapsulates the judgment result, the cable damage diagnosis result, and the power adapter identification information to generate a structured alarm data packet; The alarm data packet contains a unique event code, event level, timestamp, geographic coordinates, bite location code, and fragments of raw sensor data; S402: Implement a multi-channel early warning strategy based on event levels: First, the event level is compared with a preset threshold (for example, when the cable damage reaches the 'severe' level, or when the tearing confidence exceeds 90%, it is determined to be a high-level event). If it is a high-level event, an instant strong reminder channel is triggered, and a strong reminder notification is pushed to the user terminal's APP and an encrypted SMS message is sent through the IoT module. If it is a low-level event, a standard push notification will be sent to the user terminal's APP only through the IoT module; S403: Before sending a message, the IoT module calls the built-in lightweight decision model, combines the current network status and the historical response data of the user terminal, and dynamically selects the optimal communication protocol and data compression rate to ensure the reliability and real-time performance of alarm information in a weak network environment. S404: After receiving the alarm data packet, the user terminal parses it and displays a visual alarm on the interactive interface; The alarm interface shall at least simultaneously display a location map of the power adapter, a damage diagram of the output line, and an event timeline, and provide interactive buttons for "Handle Now", "Ignore", and "View History"; S405: The IoT module listens to user feedback commands from the user terminal and uploads the user feedback commands along with the complete data record of this event to the cloud server. This is used to update the lightweight decision model and the pet biting pattern library, thereby achieving adaptive optimization of the early warning strategy.

[0012] As a further aspect of the present invention, step S500 specifically includes the following steps: S501: In the safe mode, the control unit periodically performs a fine-grained damage self-check on the output line. This process includes: applying a series of low-voltage diagnostic signals of different frequencies and amplitudes to the output line, and simultaneously acquiring the response data of the sensor array. By analyzing the signal attenuation, phase shift and harmonic distortion characteristics, an equivalent distributed parameter circuit model of the output line in the current state is constructed. S502: Compare the constructed equivalent distributed parameter circuit model with the pre-stored benchmark model representing the cable's intact state, and calculate the offset of several key parameters, including the unit length resistance increment, insulation admittance change rate, and characteristic impedance mismatch. Based on these offsets, the damage level is quantified by a damage assessment algorithm. By analyzing the significant differences in the offsets of the equivalent circuit parameters corresponding to each sensing segment, the continuous sensing segment interval with the most drastic parameter abnormal changes is located as the interval where physical damage is most likely to occur. S503: Start the adaptive learning process: The control unit marks and stores the complete data sequence of this biting event—including the original sensor signals before the trigger, the extracted dynamic recognition features, the judgment results, the damage self-inspection results, and the user's final feedback instructions—as a learning sample package. S504: Periodically or after the number of samples accumulates to a threshold, perform incremental model optimization: retrain the pattern recognition algorithm using the newly accumulated learning sample package, wherein the weights of the training data are dynamically adjusted according to the user feedback instructions and damage self-inspection results, and samples confirmed as real bites that have caused damage are given higher weights, thereby iteratively updating the preset pet bite pattern library and the classification boundary of the pattern recognition algorithm.

[0013] As a further aspect of the present invention, step S600 specifically includes the following steps: S601: The IoT module continuously listens for feedback commands from the user terminal and parses and authenticates the commands; The feedback instructions include at least "Restore power immediately", "Maintain safe mode", "Enter learning mode" and "Adjust monitoring sensitivity"; S602: Execute the adaptive state transition and policy execution corresponding to the feedback instruction: If the instruction is "Restore power immediately", the control unit first instructs the power management unit to perform a quick re-inspection (based on sensor array) on the output line. If the re-inspection confirms that there is no continuous biting signal and the degree of damage to the cable is lower than the preset threshold, the control unit controls the power output unit to exit the safe mode, smoothly restores the power output in stages, and marks this event as "processed-restored". If the instruction is "maintain safe mode", the control unit will maintain a safe voltage output and send a detailed damage assessment report and maintenance recommendation list to the user terminal through the Internet of Things module. At the same time, the adapter work log will be marked as "awaiting manual handling". If the instruction is "Enter learning mode", the control unit temporarily adjusts the monitoring strategy of the sensor array. After the user confirms through the terminal that the pet is in a safe and controlled state, it starts data collection for a preset period of time to record the pet's typical behavior data near the adapter, in order to specifically optimize the recognition of non-destructive contact. If the instruction is "adjust monitoring sensitivity", the control unit will dynamically update the feature matching threshold in the pattern recognition algorithm according to the parameters carried by the instruction, so as to adapt to the user's specific balance requirements for false alarm rate and false negative rate. S603: Build and update a user preference model based on the user's continuous feedback history; The user preference model is used to predict the most likely user choice and automatically execute the corresponding preset strategy when the user fails to provide a clear response within a preset time in subsequent events. S604: After completing the instruction execution, the control unit generates a state transition completion report and uploads it to the cloud server via the IoT module; the report includes at least the final action taken, the key system parameters after the action is executed, and the event closed-loop classification based on user feedback, which is used to improve the adaptive learning process.

[0014] An IoT-based smart power adapter, wherein the power adapter operates using the above-described working method.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, through the deep integration of the Internet of Things (IoT) and intelligent sensing, constructs a complete closed-loop system encompassing risk perception, proactive protection, and timely alerts. First, a sensor array embedded in the output lines enables real-time monitoring and intelligent algorithm recognition, allowing for millisecond-level judgments the instant a pet bites, thus providing early warning. Second, by immediately switching to a safe power mode, it fundamentally prevents potential electric shock injuries to pets, addressing a core safety hazard. Simultaneously, the IoT module pushes event details to user terminals in real time, ensuring users are immediately informed and can handle the situation remotely, effectively solving the problem of delayed alerts. Furthermore, this method possesses self-learning and adaptive optimization capabilities, continuously accumulating data to improve recognition accuracy and adapt to user habits. Finally, the entire solution proactively and intelligently ensures pet safety, equipment safety, and power continuity without requiring physical user intervention, significantly enhancing the user experience and psychological security for pet-owning families. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating steps S100-S600 in this invention. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 A method for operating an IoT-based smart power adapter includes the following steps: S100: By using a sensor array installed on the output line of the power adapter, the physical contact signal on the output line is monitored in real time, and the physical contact signal is sent to the control unit; The sensor array includes a flexible pressure sensor, a distributed strain sensor, and an impedance measurement sensor, which are embedded in the outer sheath of the output line in a matrix configuration. S200: The control unit analyzes the physical contact signal based on a pattern recognition algorithm, extracts contact features, and compares them with a preset pet biting pattern library to identify whether it is a pet biting behavior; S300: When a pet biting behavior is detected, the control unit controls the power output unit to switch to safe mode within milliseconds; S400: The IoT module encapsulates pet biting incidents and related information into alarm messages and wirelessly sends them to the user terminal to provide user alerts; The relevant information includes the time of the event, the location of the power adapter, and the location of the bite. S500: In safe mode, a self-test program is initiated to assess the degree of damage to the output line, and the pattern recognition algorithm is updated based on historical bite data to adaptively optimize the bite recognition accuracy. S600: Based on instructions from the user terminal, adjust the operating status of the power adapter, including restoring normal power output or maintaining safe mode and triggering local alarms; This invention provides a systematic, proactive safety solution for pet-owning families through a series of closely coordinated intelligent steps: First, in terms of security protection, a fundamental shift has been achieved from passively accepting the situation to actively intervening. Existing power adapters rely solely on physical insulation or user supervision, failing to proactively protect pets in the event of an attack. This solution, however, utilizes S100's real-time monitoring and S200's intelligent algorithm recognition to accurately identify the pet's biting intent in the initial (milliseconds) stage when its teeth come into contact with the cable, rather than waiting for serious consequences such as short circuits. Subsequently, S300's safe power response mechanism, upon detecting the pet's biting behavior, intelligently switches to a safe mode, thereby eliminating the risk of electric shock or even death for the pet. This represents a safety breakthrough that related technologies have not yet achieved. Second, in terms of risk notification, a three-dimensional alarm network has been built, ranging from local silence to remote real-time synchronization. In existing related technologies, users often only know that danger has occurred after the fact when they discover damage or abnormal behavior in their pets. This solution, through the IoT alerts of S400, can not only send alarms to users' mobile phones and other terminals immediately when an event occurs, but also provide multi-dimensional information including event type, time of occurrence, precise location (such as the southeast corner of the living room), estimated level of cable damage, and even sensor data fragments at the time of triggering. This allows users to grasp the situation on the spot as soon as possible, no matter where they are, and provides accurate decision-making basis for whether to return home urgently or contact others for assistance, greatly shortening the window period of risk exposure. Third, at the level of system intelligence and reliability, a dynamic optimization mechanism with self-evolution capability has been introduced; Existing technologies often rely on fixed thresholds or simple logic, leading to false alarms (e.g., misinterpreting cable plugging / unplugging as biting) or missed detections. This solution utilizes the S500's adaptive learning and system self-checking to continuously accumulate data. Each event (regardless of whether it's actual biting) and its resulting user feedback and damage diagnosis are used to optimize the pattern recognition algorithm in the S200, enabling it to more accurately distinguish between different scenarios such as "biting," "playful scratching," and "normal plugging / unplugging," thereby continuously reducing false alarms and missed detections and improving system reliability. Simultaneously, the S600's user interaction mechanism provides the system with personalized adaptation capabilities, allowing it to learn user preferences and achieve more user-friendly control. Fourth, in terms of equipment management and user experience, it provides full-cycle services from incident handling to preventive maintenance; this solution goes beyond a single "incident response" function; through the S500's damage self-inspection and the S600's recovery mechanism, the system can automatically assess the cable health status after an incident and provide users with clear maintenance suggestions (such as "minor insulation damage, observation is recommended" or "exposed conductor, must be replaced immediately"); this helps users move from "guessing whether the equipment is safe" to "receiving clear maintenance instructions", extending the safe service life of the equipment, and may avoid more serious property damage or fire hazards by providing early warning of potential short circuit risks; Fifth, at the level of technology integration and innovation, it creates synergistic value through cross-domain technology fusion. This invention is not a simple accumulation of technologies such as sensors, controllers, and the Internet of Things, but rather achieves a deep closed-loop fusion of five major functions: perception, decision-making, execution, communication, and learning through a unique step-by-step process design. For example, sensor data drives safety response, and the response results form a user feedback closed loop through the Internet of Things. The feedback data, in turn, trains the algorithm to optimize future perception and decision-making. This organic synergy produces a "1+1>2" effect, making its overall technical effect significantly better than a solution that simply adds up the functions of each part, thus forming a highly creative and complete technical system. This invention, through the deep integration of the Internet of Things (IoT) and intelligent sensing, constructs a complete closed-loop system encompassing risk perception, proactive protection, and timely alerts. First, a sensor array embedded in the output lines enables real-time monitoring and intelligent algorithm recognition, allowing for millisecond-level judgments the instant a pet bites, thus providing early warning. Second, by immediately switching to a safe power mode, it fundamentally prevents potential electric shock injuries to pets, addressing a core safety hazard. Simultaneously, the IoT module pushes event details to user terminals in real time, ensuring users are immediately informed and can handle the situation remotely, effectively solving the problem of delayed alerts. Furthermore, this method possesses self-learning and adaptive optimization capabilities, continuously accumulating data to improve recognition accuracy and adapt to user habits. Finally, the entire solution proactively and intelligently ensures pet safety, equipment safety, and power continuity without requiring physical user intervention, significantly enhancing the user experience and psychological security for pet-owning families.

[0019] In this embodiment of the invention, step S100 specifically includes the following steps: S101: The output line is divided into multiple independent sensing segments along its length direction, and the sensor array includes a set of sensor units disposed in each sensing segment; Each sensor unit includes at least one of the flexible pressure sensors, one of the distributed strain sensors, one impedance measurement sensor, and one microcontroller node; S102: Each of the sensor units synchronously acquires the real-time physical contact signal of its sensing segment at a preset sampling frequency, wherein the real-time physical contact signal includes a pressure distribution map, strain rate of change and raw waveform data; S103: Each microcontroller node performs local preprocessing on the acquired real-time physical contact signal, extracts time-domain and frequency-domain features, and sends the preprocessed data to the control unit in real time; This invention transforms the macroscopic concept of "real-time monitoring" into an executable and highly reliable distributed sensing system; First, in terms of perceived "spatial resolution" and "event localization capability," by dividing the output cable into multiple independent sensing segments (S101) and equipping each segment with an integrated sensor unit, the system is able to locate biting behavior at the centimeter level or even more precisely for the first time. This not only informs the user that "biting has occurred," but also accurately indicates "which segment of the output cable the biting occurred on." This capability provides a crucial spatial information basis for subsequent damage assessment (S500) and targeted safety responses (such as isolating only the damaged segment as envisioned in S300), which is something that traditional adapters or integrated monitoring solutions cannot provide. Second, by integrating multi-dimensional sensors and front-end intelligent processing, signal quality and recognition reliability are improved. Each sensor unit simultaneously acquires pressure and strain signals (S102), which constitutes a "dual-channel verification" for perception. For example, simple compression (such as being pressed down by a heavy object) may generate a high-pressure signal but a low deformation rate, while biting is usually accompanied by a high deformation rate and a specific pressure pulse sequence. The microcontroller node in the sensor unit performs local preprocessing (S103), such as extracting time-domain features (peak value, rise time) and frequency-domain features (main vibration frequency). In essence, the first feature screening and noise reduction are completed at the data source. This avoids transmitting all the raw data (which may contain a lot of environmental noise) to the main control unit, thereby significantly reducing the data transmission bandwidth requirements, the computational load of the main control unit, and improving the overall system response speed. In summary, firstly, the distributed design of dividing the output line into multiple independent sensing segments, combined with the microcontroller nodes built into each sensor unit, enables precise location and parallel data acquisition of the biting behavior, providing a foundation for subsequent damage localization. Secondly, the collaborative work of the flexible pressure sensor and the distributed strain sensor can simultaneously capture the pressure distribution and cable deformation details of the contact, forming a multi-dimensional complementary physical signal, which enhances the ability to distinguish the specific contact mode of "biting" from other contact modes such as "pulling" and "squeezing". Thirdly, the signal preprocessing (extracting time and frequency domain features) performed locally in each sensing segment effectively filters out environmental noise and performs data compression before transmission, significantly reducing the data processing load of the central control unit and improving the real-time response speed of the system.

[0020] In this embodiment of the invention, step S200 specifically includes the following steps: S201: The control unit receives preprocessed data from each sensing segment and performs data fusion to generate a time-stamped matrix of the full contact status of the output line. S202: Extract at least two dynamic identification features from the full-line contact state matrix, including: tearing waveform features: calculate the impact steepness and period of the waveform based on the temporal correlation between the strain rate of change and the pressure peak. Movement pattern characteristics: Based on the propagation direction and speed of the pressure distribution map between adjacent sensing segments, identify whether it conforms to the movement pattern of tooth biting; Energy concentration characteristics: Analyze whether the energy of the sensor signal is concentrated in one or more local sensing segments within a preset time window; S203: Perform multi-level matching and judgment between the extracted dynamic recognition features and the feature vectors in the preset pet biting pattern library; First, a preliminary screening is performed to determine whether the impact steepness exceeds a first threshold and whether the energy concentration is higher than a second threshold. If the initial screening is passed, a secondary matching is performed, using a pattern recognition algorithm to calculate the similarity between the current dynamic recognition features and typical biting patterns, playful scratching patterns, and human plugging / unplugging patterns in the pet biting pattern library. S204: Based on the results of the multi-level matching judgment, if the similarity of the biting pattern output by the pattern recognition algorithm exceeds the set confidence level, and at the same time the movement pattern feature conforms to the preset biting path model, then it is finally determined to be a pet biting behavior, and a judgment result containing biting intensity and suspected pet body type is generated. First, by establishing a high-fidelity, integrated data fusion model, the problem of synchronizing and associating multi-point sensing information is solved. The "full-line contact state matrix" generated in step S201 is not a simple data summary. It aligns and fuses the pre-processed data with precise timestamps uploaded by all sensing segments in both spatial (cable position) and temporal dimensions. This creates a data structure that can reflect the overall stress and deformation state of the cable in a panoramic and continuous dynamic manner. The direct effect is that the system can not only "see" the pressure at a certain point, but also "see clearly" how a biting behavior starts from a point and how its mechanical waveform propagates and attenuates to the adjacent area. Second, by designing multi-dimensional, dynamically meaningful recognition features, the algorithm gains the ability to understand the biomechanical essence of "biting." The three types of features extracted in step S202 are a concentrated embodiment of the technical solution's creativity: Biting waveform features (impact steepness and period): directly quantify the instantaneous impact force and rhythm when teeth cut into the insulation layer, which is highly correlated with the physiological characteristics of pet chewing and can effectively filter out gentle pressure changes (such as being covered by a blanket); Movement pattern features (propagation direction and speed): mimicking the way animals "pull" or "grind" cables with their teeth. The typical trajectory of the force point moving along the cable; this feature can clearly distinguish between "destructive biting along the cable" and "repetitive playful scratching at a fixed position"; energy concentration feature: quantifies whether the destructive energy is concentrated locally, which helps to eliminate interference scenarios such as uniform pressure (such as being pressed down by a heavy object) or slight shaking of the entire cable; these features together constitute a three-dimensional "biting behavior fingerprint", which makes the subsequent classification judgment based on solid physical principles, rather than simple data pattern matching, greatly improving the algorithm's generalization ability and adaptability to pets of different sizes and breeds; Third, by designing a multi-level matching and judgment mechanism of "threshold initial screening + AI fine judgment", an engineering balance is achieved between real-time performance and accuracy. The two-level judgment strategy in step S203 is a key engineering innovation. The initial screening uses threshold comparison with minimal computational cost (such as the impact steepness threshold), which can quickly filter out more than 95% of everyday non-threatening contacts (such as slight touches or cable swaying in the wind) within microseconds, ensuring that system resources are not occupied by a large amount of meaningless data. Only suspected events that pass this "fast screening" will enter the secondary machine learning classification with higher computational cost but higher accuracy. The core benefit of this architecture design is that the system can run stably on resource-constrained embedded processors, while concentrating valuable computing resources on analyzing the most suspicious events, thereby achieving complex pattern recognition accuracy comparable to cloud computing while ensuring millisecond-level response speed. Fourth, by setting stringent and logically rigorous final judgment conditions, the risk of system malfunction is reduced to an extremely low level, establishing a cornerstone of user trust. The final judgment in step S204 is not based on the satisfaction of a single condition, but requires that "high confidence" and "compliance with the bite path model" be met simultaneously. This means that even if the machine learning model gives a high bite probability score due to a specific sample, the system will still reject it if the force movement pattern of this event does not conform to the preset bite biomechanical model. This cross-validation mechanism is like "double insurance" for the system's decision-making, greatly eliminating false triggers caused by false alarms from a single sensor, "overfitting" of the algorithm in special scenarios, or rare interference (such as strong vibrations at a specific frequency). For security protection products, avoiding frequent power cuts to important equipment due to false alarms is as important as accurately identifying real dangers. This design significantly improves the robustness of the system and the user experience, allowing users to rely on this active protection system. In summary, through the refined design of step S200, a hierarchical and highly fault-tolerant intelligent behavior recognition decision-making process was constructed. First, the data fusion in S201 generates a full-line contact state matrix, integrating scattered sensor information into a unified spatiotemporal analysis framework, laying a high-fidelity data foundation for subsequent accurate feature extraction. Second, the biting waveform features, movement pattern features, and energy concentration features defined in S202 construct a three-dimensional description of pet biting behavior from three dimensions: mechanical dynamics, spatial propagation, and energy distribution. This allows the algorithm to deeply understand the physical differences between "biting" and "playful scratching" or "human-induced insertion / removal," greatly enhancing the feature extraction accuracy. The S203 system employs a multi-level matching judgment mechanism that innovatively combines rapid threshold screening with refined machine learning classification. This not only filters out a large amount of irrelevant interference through primary thresholds to improve system real-time performance, but also performs complex pattern analysis through secondary matching based on machine learning, thus achieving an optimal balance between high-speed processing and high accuracy. Finally, the S204 system sets strict comprehensive judgment conditions, requiring both high confidence and conformity to the bite path model. This ensures that the system will only trigger critical security responses when there is sufficient evidence, effectively preventing malfunctions caused by environmental vibrations, accidental touches, and other factors, and significantly improving the reliability and user trust of the entire protection system.

[0021] In this embodiment of the invention, step S300 specifically includes the following steps: S301: After determining that a pet biting behavior has occurred, the control unit sends an emergency switching command containing biting location information to the power management unit and the power output unit. S302: The power management unit performs a millisecond-level safe mode switching operation according to the emergency switching command: Immediately disconnect the standard operating voltage and current supplied by the power output unit to the output line; S303: The control unit, in conjunction with the power management unit, applies a low voltage to the current output path of the power adapter and simultaneously switches the entire power adapter to a low-power operation state (based on the current production quality standards of the output line, it is unlikely that a common household pet will bite through the output line in one go, so the current output path of the power adapter remains effective. For a common household pet, it would take a certain amount of time to completely bite through the copper and aluminum wire cores). S304: After the safety mode is activated, based on the sensor array, the cable damage diagnosis of the output line is initiated: by injecting low-voltage test signals into both ends of the output line and combining the impedance change data fed back by the sensor array, the degree of cable damage and insulation status centered on the biting position are analyzed. S305: Based on the cable damage level, historical tearing event data, and current IoT connection status, execute adaptive security policy adjustments: If the diagnostic results indicate that the insulation layer is damaged, the power adapter will be locked in safe mode and automatic recovery will be disabled. If the damage is minor and the network connection is stable, the priority of the IoT alarm will be increased while maintaining low voltage output. Traditional leakage current protection or overcurrent protection only activates after electrical parameters are abnormal (i.e., damage has partially occurred), and the device is completely powered off after activation. The S302 is designed with a parallel dual-rail power supply architecture: the main power circuit is momentarily cut off to eliminate the source of danger, while a low voltage is applied to the current output path. The ingenuity of this design lies in the fact that it achieves the coexistence of "dangerous energy isolation" and "intelligent system continuous online operation". The sustaining voltage (such as 5V) provided by the low voltage circuit not only ensures the absolute safety of pet contact, but also provides uninterrupted "emergency power" for the "brain" (control unit) and "nerves" (sensor array) of the power adapter, so that all subsequent intelligent functions such as alarm sending (S400) and online diagnosis (S304) can continue to be executed after a safety event occurs, avoiding the "information black hole" and functional paralysis caused by the overall power outage. S304 is a rapid preliminary diagnosis for immediate strategic decision-making (S305). The online impedance diagnostic technology introduced in S304 is the key to the method's shift from "passive protection" to "proactive health management." In safe mode, by injecting a low-voltage test signal into the cable and analyzing the feedback, the system essentially performs a risk-free "electrical imaging examination" of the damaged "nerve"—the output cable. By analyzing characteristics such as impedance changes and signal attenuation, the system can construct an equivalent circuit model, thereby quantitatively distinguishing the following states: only outer skin marks, local thinning of the insulation layer but no breakdown, partial exposure of the conductor, or complete breakage. This precise assessment of the degree of damage provides users with a decision-making basis from "guessing" to "knowing." For example, the system can suggest "slightly damaged insulation layer, power reduction and observation" or "exposed conductor, must be stopped and replaced immediately," realizing the transformation from extensive scrapping to refined maintenance, improving product life and usage economy. S305's adaptive safety policy adjustment mechanism empowers the system to make dynamic decisions based on multi-dimensional real-time information. Its decision-making logic comprehensively considers: a) the objective results of damage diagnosis (e.g., locking if insulation is damaged), b) historical behavior patterns (e.g., frequent minor bites at this location), and c) the system communication status (e.g., increasing alarm frequency if network signal is good). This means the system's response is no longer fixed. For example, for a minor contact that does not cause substantial damage, if the user's home network connection is reliable, the system may choose to maintain a safe low-voltage output but not report a high-priority alarm to avoid excessively disturbing the user. Conversely, for a bite that causes serious physical damage, even if the network is interrupted, the system will lock in safe mode and attempt to issue an alarm through other backup channels (e.g., Bluetooth Mesh). This risk-assessment-based dynamic response significantly optimizes the balance between safety and user experience. The precise damage location and quantification data generated by S304 is the foundation for all subsequent intelligent services. It is not only used for immediate alarms (S400), but more importantly, it provides high-quality labeled data for the self-learning process of S500 (linking "biting behavior" with the "specific physical damage" it causes), and also guides the user recovery options of S600 (such as allowing "immediate power restoration" only when the damage is below a threshold). Therefore, step S300 is the key hub connecting the preceding and following steps in the entire solution: it receives identification instructions from S200, executes core safety operations, and generates high-value status data for all subsequent intelligent decisions (alarms, learning, recovery). In summary, by refining step S300, a multi-level proactive protection system was constructed, encompassing decision-making and execution while balancing instantaneous safety with continuous assurance. First, the command sent in S301 based on the bite location enables spatial targeting of the safety response, providing an informational basis for potential subsequent localized isolation and repair. Second, the millisecond-level dual-path switching mechanism designed in S302—immediately activating the isolated backup energy storage module safety circuit while cutting off the main dangerous power supply—not only completely eliminates the risk of electric shock to pets physically but also ensures continuous online operation of the IoT module and control unit by maintaining a safe low-voltage output, avoiding the critical problem of system "disconnection" and users not receiving alarms due to complete power failure. Furthermore, the online impedance diagnostic function introduced in S304... In safe mode, it can quantitatively assess damaged cables without risk, accurately determining the degree of insulation damage or conductor exposure through changes in electrical parameters. This transforms the vague "damaged" status into specific "damage level and location," providing users with clear and actionable maintenance decision-making basis (such as "needs immediate replacement" or "can be observed and used"). Finally, the S305's adaptive safety strategy adjustment enables the system to make intelligent decisions based on real-time diagnostic results, historical data, and network status. For example, it can lock the safety mode to prevent accidental activation when the damage is severe, or strengthen alarms when the damage is minor but the connection is stable. This optimizes resource allocation and user experience while ensuring absolute safety, achieving an efficiency upgrade from "one-time power outage protection" to "intelligent damage management and continuous risk monitoring."

[0022] In this embodiment of the invention, step S400 specifically includes the following steps: S401: The control unit encapsulates the judgment result, the cable damage diagnosis result, and the power adapter identification information to generate a structured alarm data packet; The alarm data packet contains a unique event code, event level, timestamp, geographic coordinates, bite location code, and fragments of raw sensor data; S402: Implement a multi-channel early warning strategy based on event levels: First, the event level is compared with a preset threshold (for example, when the cable damage reaches the 'severe' level, or when the tearing confidence exceeds 90%, it is determined to be a high-level event). If it is a high-level event, an instant strong reminder channel is triggered, and a strong reminder notification is pushed to the user terminal's APP and an encrypted SMS message is sent through the IoT module. If it is a low-level event, a standard push notification will be sent to the user terminal's APP only through the IoT module; S403: Before sending a message, the IoT module calls the built-in lightweight decision model, combines the current network status and the historical response data of the user terminal, and dynamically selects the optimal communication protocol and data compression rate to ensure the reliability and real-time performance of alarm information in a weak network environment. S404: After receiving the alarm data packet, the user terminal parses it and displays a visual alarm on the interactive interface; The alarm interface shall at least simultaneously display a location map of the power adapter, a damage diagram of the output line, and an event timeline, and provide interactive buttons for "Handle Now", "Ignore", and "View History"; S405: The IoT module listens to user feedback commands from the user terminal and uploads the user feedback commands along with the complete data record of this event to the cloud server to update the lightweight decision model and the pet biting pattern library, thereby achieving adaptive optimization of the early warning strategy. Step S400 constructs an alarm and decision support system that integrates information integration, intelligent distribution, reliable transmission, intuitive interaction, and closed-loop learning. It changes the outdated mode of traditional devices simply sending "fault codes" and upgrades an alarm event into a key interactive node that drives user action and optimizes system performance. Specifically, this is reflected in the following five levels of progress: The first layer: Through structured and standardized data encapsulation, a qualitative change has been achieved in alarm information from "data fragments" to "traceable evidence chains"; S401 requires the generation of structured alarm data packets containing elements such as a unique event code, geographic coordinates, and raw sensor data fragments. This is far more than a simple information package; its core value lies in: traceability and accountability: unique codes and timestamps give each event a unique ID in the server log, facilitating later querying, analysis, and accountability determination; preserving "original evidence" for advanced analysis: including raw data fragments (such as pressure waveform fragments) allows cloud-based or expert analysis, algorithm optimization, or dispute arbitration to trace back to the most basic sensor signals, rather than relying solely on the initial processing results at the device end, thus improving data credibility and secondary development value; a common language for inter-system interconnection: the standardized structure allows the alarm packet to seamlessly interface with smart home platforms, property management systems, or pet insurance service platforms, providing a plug-and-play data interface for cross-system linkage (such as triggering smart camera rotation). The second layer: Through a multi-channel intelligent distribution strategy based on event levels, information push has evolved from "indiscriminate bombardment" to "context-aware and precise delivery." The S402's multi-channel early warning strategy introduces key risk classification logic: For "high-level events" (such as diagnosing conductor exposure), the system simultaneously activates a strong APP reminder (which can bypass the phone's silent mode) and SMS message. This "double insurance" design takes into account situations where the user may not have enabled APP notification permissions or the phone is in Do Not Disturb mode, maximizing the guarantee that life-threatening alarms can be perceived by the user in a timely manner; For "low-level events" (such as only skin indentation), only APP push is used to avoid unnecessary interference. This classification mechanism reflects respect for the user's context and significantly reduces "alarm fatigue" caused by frequent false alarms or low-value alarms, making users more willing to keep the system online for a long time. The third layer: Through a forward-looking network adaptive mechanism, it overcomes the "last mile" reliability challenge for IoT devices in real, complex network environments. The S403's lightweight decision model is the core technology ensuring alarm effectiveness. Traditional solutions often assume a consistently good network, leading to alarm loss in signal dead zones. This solution's model can: dynamically select protocols: using TCP to ensure data integrity when Wi-Fi signals are strong; automatically switching to more loss-resistant UDP or low-power Bluetooth broadcast when signals are weak; intelligently compress data: dynamically adjusting the data compression rate based on network quality, prioritizing the sending of core text alarms when the network is extremely poor, and supplementing with detailed images or waveforms when the network is good; predictive retransmission: predictively retransmitting multiple times for critical high-level events based on historical connection success rates. This ensures that alarm delivery rate (reliability) and real-time performance are systematically guaranteed, regardless of signal weakness in areas such as basements or corners, which is the foundation for achieving practical value. The fourth layer: Through a highly visualized interactive interface, professional technical data is transformed into actionable guidelines that are understandable and operable for ordinary users; the S404's visualized alarm interface design aims to solve the pain point of "users receiving alarms but not knowing what to do"; damage diagram: intuitively showing which part of the cable is damaged and the type of damage (indentation / breakage / fracture), replacing the professional description of "insulation resistance dropped to X megohms"; location map: clearly indicating that it is the adapter next to the sofa in the living room, not the one in the study; clear handling buttons: providing options such as "handle immediately" and "ignore" to directly guide user action; this design greatly reduces the user's understanding cost and decision-making threshold, transforming the user from a passive information receiver to an active problem solver, improving handling efficiency and system usability; The fifth layer: By constructing a reinforcement learning closed loop of "user feedback - cloud optimization," the alarm system possesses the vitality for continuous evolution. The S405's feedback upload and model optimization mechanism is the ultimate manifestation of system intelligence. Every user click (such as clicking "ignore" or "handle immediately") is regarded by the system as a manual annotation of the accuracy and importance of this alarm. If the user frequently "ignores" a certain pattern of low-damage alarms, the cloud model will learn to raise the recognition threshold of such events, reducing subsequent false alarms. If the user always "handles immediately" a certain type of high-damage event, the system will consolidate the relevant features to ensure that similar events in the future can be identified with high confidence and trigger strong reminders. This makes the entire alarm strategy no longer a fixed setting at the factory, but a personalized system that can learn the specific family pet's behavior habits and user preferences, constantly optimizing itself, becoming more accurate and more considerate with use. In summary, step S400 constructs a highly reliable, user-friendly, and self-optimizing intelligent information interaction hub. It upgrades a simple "sending notification" into a complete "decision support and system evolution" event. Specifically, the structured alarm data packet design in S401, through unified encoding and inclusion of original data fragments, not only ensures the integrity and traceability of information but also provides a standardized data foundation for backend big data analysis and responsibility delineation. S402's multi-channel early warning strategy based on event levels innovatively achieves risk classification management, ensuring that high-urgency events (such as conductor exposure) can penetrate the user's attention barrier through strong alert channels (such as APP push notifications and SMS messages), while low-level events avoid excessive disturbance, significantly improving the intelligence of the alarm system and the rationality of the user experience. The lightweight decision model introduced in S403 can dynamically optimize communication strategies based on real-time network conditions and the user's terminal historical behavior. This mechanism fundamentally guarantees the reliability of Wi-Fi- Even in remote corners of the home with weak Fi signals or unstable mobile networks, critical alarms can still be delivered with high probability and low latency, greatly enhancing the system's reliability in real-world, complex environments. S404 provides a visual alarm interface on the user terminal, transforming abstract sensor data and event codes into intuitive maps, diagrams, and timelines, and giving users clear action options. This significantly lowers the barrier to understanding and operation for users, enabling even non-professional users to quickly and accurately grasp the situation and make effective decisions. Finally, the feedback loop and cloud learning mechanism established by S405 ensures that every user interaction becomes a source of optimization for the system, continuously training the warning model and the bite pattern library, achieving adaptive growth of alarm strategies from "preset and fixed" to "more accurate with use." Therefore, step S400 not only solves the basic problem of "timely reminders" but also establishes a deeply collaborative and constantly evolving security information ecosystem with users through intelligent hierarchical management, reliable transmission, intuitive interaction, and continuous learning.

[0023] In this embodiment of the invention, step S500 specifically includes the following steps: S501: In the safe mode, the control unit periodically performs a fine-grained damage self-check on the output line. This process includes: applying a series of low-voltage diagnostic signals of different frequencies and amplitudes to the output line, and simultaneously acquiring the response data of the sensor array. By analyzing the signal attenuation, phase shift and harmonic distortion characteristics, an equivalent distributed parameter circuit model of the output line in the current state is constructed. S502: Compare the constructed equivalent distributed parameter circuit model with the pre-stored benchmark model representing the cable's intact state, and calculate the offset of several key parameters, including the unit length resistance increment, insulation admittance change rate, and characteristic impedance mismatch. Based on these offsets, the damage level is quantified by a damage assessment algorithm. By analyzing the significant differences in the offsets of the equivalent circuit parameters corresponding to each sensing segment, the continuous sensing segment interval with the most drastic parameter abnormal changes is located as the interval where physical damage is most likely to occur. S503: Start the adaptive learning process: The control unit marks and stores the complete data sequence of this biting event—including the original sensor signals before the trigger, the extracted dynamic recognition features, the judgment results, the damage self-inspection results, and the user's final feedback instructions—as a learning sample package. S504: Periodically or after the number of samples has accumulated to a threshold, perform incremental model optimization: retrain the pattern recognition algorithm using the newly accumulated learning sample package, wherein the weights of the training data are dynamically adjusted according to the user feedback instructions and damage self-inspection results, and samples that are confirmed to be real bites and have caused damage are given higher weights, thereby iteratively updating the preset pet bite pattern library and the classification boundary of the pattern recognition algorithm. Step S500 endows the entire system with the dual effects of "self-check-up" and "self-evolution," transforming an isolated security event into a valuable data asset that drives long-term system performance improvement and enables predictive maintenance. This is not only about optimizing subsequent identification, but also profoundly reflected in the following three dimensions: First, at the level of damage diagnosis, it achieves a paradigm shift from "qualitative continuity testing" to "quantitative pathological analysis," opening up the possibility of predictive maintenance. Traditional "self-testing" solutions are usually limited to detecting whether the cable is completely open or short-circuited, a crude, either-or judgment. S501 is a periodic, in-depth diagnostic tool used for quantitative assessment and learning. Compared to S304, S501's monitoring is more detailed. The refined damage self-testing defined by S501-S502 works more like a precise "electrical endoscopy" inspection. By injecting a series of low-voltage test signals of different frequencies into the cable and analyzing its response, the system can construct an equivalent distributed parameter circuit model reflecting the cable's current complete electrical characteristics. By comparing this model with the "health baseline model" at the factory, it can quantitatively calculate: the increase in resistance per unit length, indicating that the conductor may have micro-cracks or localized damage due to biting. Thinning leads to increased resistance and a higher risk of overheating; the rate of change in insulation admittance directly reflects the degree of aging, moisture absorption, or damage to the insulation layer. Increased admittance means decreased insulation performance and an increased risk of leakage current; the mismatch of characteristic impedance indicates that the uniformity of the cable is disrupted, which will affect the transmission of high-frequency signals and accurately locate the physical area where damage is most likely to occur; this diagnostic depth allows the system to distinguish between different levels of damage states such as "skin scratches", "moderate wear of the insulation layer but no breakdown", "slight exposure of the conductor" and "imminent breakage". It can issue early warnings and provide differentiated maintenance recommendations (such as "replace next week" or "can continue to be used but avoid bending") before potential faults (such as overheating due to increased resistance or leakage due to insulation deterioration) actually cause serious accidents, thus realizing a leap from passive maintenance of "replacing when it breaks" to "predictive health management"; Second, in terms of data value, a closed loop for the generation and extraction of "event records" to "high-quality labeled learning samples" has been constructed, solving the core problem of data quality in machine learning in real-world scenarios. Many intelligent systems can collect data, but the data is often messy and lacks effective annotation, leading to inefficient model optimization. S503-S504 designs a meticulously crafted data pipeline that encapsulates the complete context of an event—including the initial signal before triggering (“cause”), the algorithm's judgment and response (“process”), the precise damage quantification result (“effect”), and the user's final action feedback (“human verification”)—into a structured learning sample package. The key innovation lies in the quality and dimensionality of the “annotation”: samples are not only labeled as “bite” or “non-bite”, but also associated with precise damage quantification indicators (such as a 5% change in insulation admittance). The system considers both the user's value judgment (whether the user considers the alert "important" or "annoying"); this provides extremely rich and accurate monitoring signals for algorithm optimization; dynamic weight adjustment mechanism: the system assigns different weights to samples based on the damage results and user feedback. A bite sample that is confirmed by the user and causes severe conductor exposure has the highest weight; a "suspected" sample that is frequently ignored by the user and diagnosed as having no damage is marked as a possible false alarm to help the algorithm converge and avoid learning noise. This process ensures that the data used by the system for learning is "golden data" verified in the real world, improving the efficiency and correctness of model optimization. Third, at the system evolution level, it achieves continuous alignment between algorithm performance and real-world user scenarios and preferences, ensuring long-term practical value. Based on the aforementioned high-quality samples, the incremental optimization of the S504 model is not blind; it enables the core pattern recognition algorithm to: (The sentence is incomplete in the original text.) Adapt to individualized environments: Learn the unique biting habits of pets in a particular household (e.g., a cat likes to gently nibble on cable ends) and common sources of disturbance in the household (e.g., a robot vacuum cleaner regularly bumping into cables). Converging to an acceptable risk threshold for users: By analyzing user feedback on various events, the system can dynamically adjust its recognition sensitivity, finding the optimal balance between "not missing any dangers" and "not generating too many false alarms" that aligns with the user's preferences and risk tolerance. This makes the product no longer fixed at the factory, but a "living" system that becomes more accurate and better suited to the actual needs of the family as the user goes through the process. This ability to continuously evolve constitutes a technological barrier that is difficult to imitate and a lasting advantage in user experience. In summary, step S500 expands the system's capabilities from "single event response" to "full lifecycle health management" and "continuous intelligent evolution." By introducing refined electrical diagnostics and a data-driven learning loop, it achieves three fundamental improvements: First, its defined refined damage self-inspection (S501-S502) completely surpasses traditional simple continuity detection. By applying multi-frequency diagnostic signals and constructing an equivalent distributed parameter circuit model, it can non-invasively quantify and analyze the internal state of cables (such as micro-damage to the insulation layer and early-stage conductor corrosion), thereby achieving accurate diagnosis from "whether it is damaged" to "the degree and mode of damage," providing users with a basis for predictive maintenance decisions and greatly improving the scientific nature and safety of equipment management; Second, its established adaptive learning process (S503-S504) creates a... A complete data value loop is established by encapsulating the raw sensor data, judgment results, damage reports, and user feedback for each event into labeled learning sample packages. The classification model is then incrementally optimized using dynamic weights based on feedback results. This allows the system's recognition algorithm to continuously learn from real-world scenarios, constantly correcting misjudgments and strengthening correct judgments, significantly improving the system's long-term accuracy and adaptability to the unique behaviors of different family pets. Ultimately, this mechanism enables the entire system to autonomously evolve from experience. Its model optimization process is directly linked to core safety and user experience indicators (such as user-accepted false alarms, missed alarms, and damage consequences), ensuring the continuous growth of technological utility. This transforms the power adapter from a static, fixed-function hardware product into an intelligent safety terminal that becomes increasingly accurate and reliable with use.

[0024] In this embodiment of the invention, step S600 specifically includes the following steps: S601: The IoT module continuously listens for feedback commands from the user terminal and parses and authenticates the commands; The feedback instructions include at least "Restore power immediately", "Maintain safe mode", "Enter learning mode" and "Adjust monitoring sensitivity"; S602: Execute the adaptive state transition and policy execution corresponding to the feedback instruction: If the instruction is "Restore power immediately", the control unit first instructs the power management unit to perform a quick re-inspection (based on sensor array) on the output line. If the re-inspection confirms that there is no continuous biting signal and the degree of damage to the cable is lower than the preset threshold, the control unit controls the power output unit to exit the safe mode, smoothly restores the power output in stages, and marks this event as "processed-restored". If the instruction is "maintain safe mode", the control unit will maintain a safe voltage output and send a detailed damage assessment report and maintenance recommendation list to the user terminal through the Internet of Things module. At the same time, the adapter work log will be marked as "awaiting manual handling". If the instruction is "Enter learning mode", the control unit temporarily adjusts the monitoring strategy of the sensor array. After the user confirms through the terminal that the pet is in a safe and controlled state, it starts data collection for a preset period of time to record the pet's typical behavior data near the adapter, in order to specifically optimize the recognition of non-destructive contact. If the instruction is "adjust monitoring sensitivity", the control unit will dynamically update the feature matching threshold in the pattern recognition algorithm according to the parameters carried by the instruction, so as to adapt to the user's specific balance requirements for false alarm rate and false negative rate. S603: Build and update a user preference model based on the user's continuous feedback history; The user preference model is used to predict the most likely user choice and automatically execute the corresponding preset strategy when the user fails to provide a clear response within a preset time in subsequent events. S604: After completing the instruction execution, the control unit generates a state transition completion report and uploads it to the cloud server via the IoT module; the report includes at least the final action taken, the key system parameters after the action is executed, and the event closed-loop classification based on user feedback, which is used to improve the adaptive learning process; Step S600 transforms the user from an external operator into a core participant and definer of the system's internal intelligent closed loop. Through a sophisticated set of interaction, learning, and prediction mechanisms, it enables the system to dynamically adapt to the user's personalized needs and behavioral patterns, thereby achieving a fundamental leap from "standardized security protection" to "personalized security services," which is reflected in the following four progressive levels: First, at the level of interactive security, a protection upgrade has been established from "simple command execution" to "security intent interpretation and verification." The core intelligence of the S601-S602 command parsing and execution mechanism lies in the fact that it does not blindly execute user commands. Taking the "immediately restore power" command as an example, the system will not directly close the circuit breaker, but will first trigger a rapid re-check (S602). This step is crucial, as it checks two points: 1) whether there is still a continuous biting signal (to prevent power from being restored while the pet is still biting); 2) whether the cable damage is below the safety threshold (to prevent power from being supplied while the conductor is exposed). Only when both conditions are met will the system execute "phased smooth restoration." This means that even if the user issues an unsafe command due to negligence or anxiety, the system retains a final safety logic check based on objective sensor data, preventing secondary damage that may be caused by human error. This design, while giving the user control, embeds an unavoidable safety protection, achieving safety redundancy in human-machine collaboration. Secondly, at the level of system personalization, a new paradigm of "guided learning" has been created, in which users directly participate in model training. The "enter learning mode" command in S602 is a breakthrough design. Traditional machine learning relies on passively collected data, which may contain a large number of unlabeled or fuzzy samples. However, "learning mode" allows users to actively and purposefully allow the system to collect data in specific scenarios when they confirm that their pet is safe and under control (such as holding the pet or using a toy to guide it). For example, users can guide their pet to play and scratch the cable near the adapter (but not bite it), thereby generating a large number of clear "non-destructive contact" negative samples; or simulate human plugging and unplugging actions. This high-quality data, actively created and labeled by users, can "teach" the system to distinguish between normal family activities and real threats in an extremely efficient manner, greatly accelerating the algorithm's adaptation to specific family environments, solving the problem of high false alarm rate in the early stage of cold start, and achieving a simultaneous and rapid improvement in user experience and system performance. Third, at the level of system autonomy, the evolution from "responsive service" to "predictive service" has been achieved by building a user preference model. The value of the user preference model built by S603 lies in handling the common but critical scenario of "user non-response." By analyzing the user's historical feedback (for example, the user chose "keep safe mode and wait for inspection" in the past three similar events), the system can build a personalized decision profile. When a new event occurs and the user does not respond within the specified time, the system no longer simply waits or sends repeated reminders, but can predict the user's most likely choice based on the preference model and automatically execute preset strategies. For example, for users who always act cautiously, the system may automatically keep safe mode and schedule a more detailed remote diagnostic; for users who are often away from home but trust the system, it may automatically restore power and send a summary report in case of minor damage. This predictive service significantly improves the system's availability and user experience, making it a considerate and user-aware smart butler, rather than a tool that requires constant attention. Fourth, at the system evolution level, user interaction behavior is solidified into a high-quality data source driving global optimization. The state transition completion report generated by S604 goes far beyond mere recording. It structurally encapsulates and uploads the complete outcome of a user interaction—the user command, the system's execution action, and the system state after execution. This report is the final closed-loop key of the aforementioned "learning sample package," and the historical response data of the user terminal in S403 is based on the S604 report. It clearly tells the learning algorithm: "For a certain identified event, the user's final handling method is A, and the system state after handling is B." This makes the algorithm's optimization goal not only include "accurate identification" but also the higher-level task of "guiding users to complete efficient and safe handling." For example, if a certain alarm mode always leads users to choose "ignore" and subsequent diagnosis shows no damage, the algorithm will learn to adjust the generation rules of such alarms. Conversely, if a certain minor damage alarm always prompts users to perform preventative maintenance, it proves that the alarm is valuable. In this way, the optimization direction of the entire system is strongly aligned with the user's actual value judgment and handling effect, ensuring that the system evolution always moves in the direction of improving usability and user satisfaction. In summary, step S600 constructs an ultimate closed loop of two-way intelligent interaction and personalized adaptive management, elevating the system from a passive command execution tool to an intelligent partner capable of understanding user intent, learning user habits, and proactively providing collaborative services. Specifically, the adaptive state transition mechanism defined in S601-S602 parses high-level user commands such as "immediately restore power" into a set of safe execution sequences including safety re-checks, phased recovery, and data marking. While granting users ultimate control, it ensures that any recovery operation undergoes systematic safety verification, eliminating secondary risks caused by user misoperation. The innovative commands such as "learning mode" and "sensitivity adjustment" introduced in S602 allow users to actively guide system optimization for the first time. By providing positive and negative samples in a safe environment or directly adjusting thresholds, users are deeply involved in the personalized customization of the algorithm, thereby greatly improving the system's adaptability to different complex home environments and the user's sense of control over the system. The user preference model constructed in S603 analyzes... The system leverages historical user feedback data to intelligently predict and execute preset strategies that best suit user habits (such as favoring "keeping in safe mode" pending inspection) when users fail to respond promptly. This represents a leap from "waiting for instructions" to "predictive service," ensuring that the system can still make reasonable and safe automated decisions even when users cannot handle the situation immediately. Finally, the state transition completion report generated by S604 standardizes and uploads the results of each interaction. This not only provides clear audit logs for users and the cloud but, more importantly, provides high-quality training data with feedback for the adaptive learning process of claim 6. This transforms users' on-site decisions into the core fuel driving the continuous evolution of the entire system. Therefore, claim 7, by endowing the system with deep interactivity, personalized adaptability, and predictive execution capabilities based on user behavior, ultimately completes the entire closed loop of the intelligent protection solution: "perception-decision-execution-interaction-evolution," making it a truly user-centric, increasingly intelligent collaborative security system.

[0025] An IoT-based smart power adapter, wherein the power adapter operates using the above-described working method.

[0026] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for operating a smart power adapter based on the Internet of Things, characterized in that, Includes the following steps: S100: By using a sensor array installed on the output line of the power adapter, the physical contact signal on the output line is monitored in real time, and the physical contact signal is sent to the control unit; The sensor array includes a flexible pressure sensor, a distributed strain sensor, and an impedance measurement sensor, which are embedded in the outer sheath of the output line in a matrix configuration. S200: The control unit analyzes the physical contact signal based on a pattern recognition algorithm, extracts contact features, and compares them with a preset pet biting pattern library to identify whether it is a pet biting behavior; S300: When a pet biting behavior is detected, the control unit controls the power output unit to switch to safe mode within milliseconds; S400: The IoT module encapsulates pet biting incidents and related information into alarm messages and wirelessly sends them to the user terminal to provide user alerts; The relevant information includes the time of the event, the location of the power adapter, and the location of the bite. S500: In safe mode, a self-test program is initiated to assess the degree of damage to the output line, and the pattern recognition algorithm is updated based on historical bite data to adaptively optimize the bite recognition accuracy. S600: Based on instructions from the user terminal, adjust the operating status of the power adapter, including restoring normal power output or maintaining a safe mode and triggering a local alarm.

2. The working method of an IoT-based smart power adapter according to claim 1, characterized in that, Step S100 specifically includes the following steps: S101: The output line is divided into multiple independent sensing segments along its length direction, and the sensor array includes a set of sensor units disposed in each sensing segment; Each sensor unit includes at least one of the flexible pressure sensors, one of the distributed strain sensors, one impedance measurement sensor, and one microcontroller node; S102: Each of the sensor units synchronously acquires the real-time physical contact signal of its sensing segment at a preset sampling frequency, wherein the real-time physical contact signal includes a pressure distribution map, strain rate of change and raw waveform data; S103: Each microcontroller node performs local preprocessing on the acquired real-time physical contact signal, extracts time-domain and frequency-domain features, and sends the preprocessed data to the control unit in real time.

3. The working method of an IoT-based smart power adapter according to claim 2, characterized in that, Step S200 specifically includes the following steps: S201: The control unit receives preprocessed data from each sensing segment and performs data fusion to generate a time-stamped matrix of the full contact status of the output line. S202: Extract at least two dynamic identification features from the full-line contact state matrix, including: tearing waveform features: calculate the impact steepness and period of the waveform based on the temporal correlation between the strain rate of change and the pressure peak. Movement pattern characteristics: Based on the propagation direction and speed of the pressure distribution map between adjacent sensing segments, identify whether it conforms to the movement pattern of tooth biting; Energy concentration characteristics: Analyze whether the energy of the sensor signal is concentrated in one or more local sensing segments within a preset time window; S203: Perform multi-level matching and judgment between the extracted dynamic recognition features and the feature vectors in the preset pet biting pattern library; First, a preliminary screening is performed to determine whether the impact steepness exceeds a first threshold and whether the energy concentration is higher than a second threshold. If the initial screening is passed, a secondary matching is performed, using a pattern recognition algorithm to calculate the similarity between the current dynamic recognition features and typical biting patterns, playful scratching patterns, and human plugging / unplugging patterns in the pet biting pattern library. S204: Based on the results of the multi-level matching judgment, if the similarity of the biting pattern output by the pattern recognition algorithm exceeds the set confidence level, and at the same time the movement pattern feature conforms to the preset biting path model, then it is finally determined to be a pet biting behavior, and a judgment result containing the biting intensity and the suspected pet body type is generated.

4. The working method of an IoT-based smart power adapter according to claim 3, characterized in that, Step S300 specifically includes the following steps: S301: After determining that a pet biting behavior has occurred, the control unit sends an emergency switching command containing biting location information to the power management unit and the power output unit. S302: The power management unit performs a millisecond-level safe mode switching operation according to the emergency switching command: Immediately disconnect the standard operating voltage and current supplied by the power output unit to the output line; S303: The control unit, in conjunction with the power management unit, applies a low voltage to the current output path of the power adapter and simultaneously switches the entire power adapter to a low-power operating state. S304: After the safety mode is activated, based on the sensor array, the cable damage diagnosis of the output line is initiated: by injecting low-voltage test signals into both ends of the output line and combining the impedance change data fed back by the sensor array, the degree of cable damage and insulation status centered on the biting position are analyzed. S305: Based on the cable damage level, historical tearing event data, and current IoT connection status, execute adaptive security policy adjustments: If the diagnostic results indicate that the insulation layer is damaged, the power adapter will be locked in safe mode and automatic recovery will be disabled. If the damage is minor and the network connection is stable, the priority of the IoT alarm should be increased while maintaining low voltage output.

5. The working method of an IoT-based smart power adapter according to claim 4, characterized in that, Step S400 specifically includes the following steps: S401: The control unit encapsulates the judgment result, the cable damage diagnosis result, and the power adapter identification information to generate a structured alarm data packet; The alarm data packet contains a unique event code, event level, timestamp, geographic coordinates, bite location code, and fragments of raw sensor data; S402: Implement a multi-channel early warning strategy based on event levels: First, the event level is compared with a preset threshold. If it is a high-level event, an instant strong reminder channel is triggered. The strong reminder notification is pushed to the user terminal's APP through the Internet of Things module and an encrypted SMS message is sent. If it is a low-level event, a standard push notification will be sent to the user terminal's APP only through the IoT module; S403: Before sending a message, the IoT module calls the built-in lightweight decision model, combines the current network status and the historical response data of the user terminal, and dynamically selects the optimal communication protocol and data compression rate to ensure the reliability and real-time performance of alarm information in a weak network environment. S404: After receiving the alarm data packet, the user terminal parses it and displays a visual alarm on the interactive interface; The alarm interface shall at least simultaneously display a location map of the power adapter, a damage diagram of the output line, and an event timeline, and provide interactive buttons for "Handle Now", "Ignore", and "View History"; S405: The IoT module listens to user feedback commands from the user terminal and uploads the user feedback commands along with the complete data record of this event to the cloud server. This is used to update the lightweight decision model and the pet biting pattern library, thereby achieving adaptive optimization of the early warning strategy.

6. The working method of an IoT-based smart power adapter according to claim 5, characterized in that, Step S500 specifically includes the following steps: S501: In the safe mode, the control unit periodically performs a fine-grained damage self-check on the output line. This process includes: applying a series of low-voltage diagnostic signals of different frequencies and amplitudes to the output line, and simultaneously acquiring the response data of the sensor array. By analyzing the signal attenuation, phase shift and harmonic distortion characteristics, an equivalent distributed parameter circuit model of the output line in the current state is constructed. S502: Compare the constructed equivalent distributed parameter circuit model with the pre-stored benchmark model representing the cable's intact state, and calculate the offset of several key parameters, including the unit length resistance increment, insulation admittance change rate, and characteristic impedance mismatch. Based on these offsets, the damage level is quantified by a damage assessment algorithm. By analyzing the significant differences in the offsets of the equivalent circuit parameters corresponding to each sensing segment, the continuous sensing segment interval with the most drastic parameter abnormal changes is located as the interval where physical damage is most likely to occur. S503: Start the adaptive learning process: The control unit marks and stores the complete data sequence of this biting event—including the original sensor signals before the trigger, the extracted dynamic recognition features, the judgment results, the damage self-inspection results, and the user's final feedback instructions—as a learning sample package. S504: Periodically or after the number of samples accumulates to a threshold, perform incremental model optimization: retrain the pattern recognition algorithm using the newly accumulated learning sample package, wherein the weights of the training data are dynamically adjusted according to the user feedback instructions and damage self-inspection results, and samples confirmed as real bites that have caused damage are given higher weights, thereby iteratively updating the preset pet bite pattern library and the classification boundary of the pattern recognition algorithm.

7. The working method of an IoT-based smart power adapter according to claim 6, characterized in that, Step S600 specifically includes the following steps: S601: The IoT module continuously listens for feedback commands from the user terminal and parses and authenticates the commands; The feedback instructions include at least "Restore power immediately", "Maintain safe mode", "Enter learning mode" and "Adjust monitoring sensitivity"; S602: Execute the adaptive state transition and policy execution corresponding to the feedback instruction: If the instruction is "Restore power immediately", the control unit first instructs the power management unit to perform a quick re-inspection on the output line. If the re-inspection confirms that there is no continuous biting signal and the degree of damage to the cable is lower than the preset threshold, the control unit controls the power output unit to exit the safe mode, smoothly restores the power output in stages, and marks this event as "processed-restored". If the instruction is "maintain safe mode", the control unit will maintain a safe voltage output and send a detailed damage assessment report and maintenance recommendation list to the user terminal through the Internet of Things module, while marking the adapter work log as "awaiting manual handling". If the instruction is "Enter learning mode", the control unit temporarily adjusts the monitoring strategy of the sensor array. After the user confirms through the terminal that the pet is in a safe and controlled state, it starts data collection for a preset period of time to record the pet's typical behavior data near the adapter, in order to specifically optimize the recognition of non-destructive contact. If the instruction is "adjust monitoring sensitivity", the control unit will dynamically update the feature matching threshold in the pattern recognition algorithm according to the parameters carried by the instruction, so as to adapt to the user's specific balance requirements for false alarm rate and false negative rate. S603: Build and update a user preference model based on the user's continuous feedback history; The user preference model is used to predict the most likely user choice and automatically execute the corresponding preset strategy when the user fails to provide a clear response within a preset time in subsequent events. S604: After completing the instruction execution, the control unit generates a state transition completion report and uploads it to the cloud server via the IoT module; the report includes at least the final action taken, the key system parameters after the action is executed, and the event closed-loop classification based on user feedback, which is used to improve the adaptive learning process.

8. A smart power adapter based on the Internet of Things, characterized in that, The power adapter is operated using the method described in any one of claims 1-7.

Images