An abnormal behavior recognition method and system based on a smart door lock

Abstract

The application relates to the technical field of intelligent door locks, and discloses an abnormal behavior recognition method and system based on an intelligent door lock, which comprises an electric parameter acquisition module, a decoupling loop analysis module, a parameter correction module and a power protection control module. The operation method is as follows: firstly, voltage and current sequences are synchronously acquired during continuous authentication failure, a starting charge pulse of a discrete event is recognized, a decoupling recovery limited index is inverted, and an equivalent recovery time constant is derived; then, the constant is used to construct a time domain filter-out mask and an energy backfill deduction term, dynamic interference compensation is performed on a voltage recovery hysteresis rate and a cumulative energy consumption slope, so that detection errors caused by power network recovery lag are eliminated; finally, the compensated parameters are input into a particle swarm optimization model to calculate a battery health degree score, the score is compared with a threshold value, continuous authentication failure is accurately distinguished from illegal intrusion or battery performance limitation, and an alarm or an emergency interface auxiliary authentication strategy is executed accordingly.

Description

Technical Field

[0001] This invention relates to the field of smart door lock technology, and more specifically, to a method and system for identifying abnormal behavior based on smart door locks. Background Technology

[0002] With the development of IoT technology, smart locks have been widely used in homes, hotels, and offices. Smart locks typically integrate multiple functions such as biometric recognition, password input, and wireless communication. Since smart locks mostly use dry cell batteries or lithium batteries for power, the stability of their power management system and the monitoring of battery health are crucial for the normal use and security of the lock. In practical applications, smart locks often face situations of continuous authentication failures. This can be caused by two distinct reasons: first, normal user error or natural battery degradation leading to an inability to drive the load; second, malicious unauthorized intrusion, such as wireless flooding attacks or brute-force continuous trial and error, causing the system to frequently wake up and execute high-power operations within a short period.

[0003] In existing technologies, the identification of abnormal behavior and the detection of battery health are usually performed independently. For battery monitoring, a static voltage threshold method is mostly used, i.e., an alarm is triggered when the voltage is detected to be below a set value; for abnormal intrusion, only the number of consecutive failures is counted to trigger a lockout or alarm. However, this simple processing logic has significant technical shortcomings when facing high-frequency continuous authentication failure scenarios: The inability to distinguish between transient fatigue in the power supply circuit and battery aging: During consecutive high-frequency certification failures, the system load exhibits high-density pulse characteristics. At this time, the decoupling capacitors in the power distribution network need to be repeatedly charged and discharged to maintain voltage stability. If the pulse interval is too short, the decoupling circuit cannot fully recover in time, leading to a false rapid drop in the power supply voltage, while the current loop generates dynamic interference. Existing technology cannot identify this voltage recovery lag caused by the power network's delayed recovery, often misdiagnosing it as battery aging or insufficient charge.

[0004] High false positive rate in abnormal behavior identification: Due to the inability to isolate interference introduced by limited power network recovery, the system struggles to distinguish between high-frequency loads caused by external unauthorized intrusion and authentication failures due to limited battery performance based on electrical parameters. This can lead to false alarms of battery failure when under attack, or false alarms of system attack when battery performance deteriorates due to low temperature or aging, thereby reducing the reliability of the security system and the user experience.

[0005] Lack of dynamic interference compensation mechanism: Existing power protection strategies are often one-size-fits-all, lacking adaptive compensation based on the circuit's dynamic recovery capability. When the battery has not yet truly aged but its power supply capacity is temporarily reduced due to high-frequency operation, there is a lack of effective algorithms to assess the true battery health score, making it impossible to implement a graded power protection strategy. Summary of the Invention

[0006] This invention provides a method and system for identifying abnormal behavior based on smart door locks, which solves the technical problems mentioned in the background art.

[0007] In the first aspect, an abnormal behavior recognition system based on a smart door lock includes an electrical parameter acquisition module, a decoupling loop analysis module, a parameter correction module, and a power protection control module. The electrical parameter acquisition module is configured to simultaneously acquire the battery terminal voltage sequence and the loop current sequence during continuous authentication failures of the smart door lock. The decoupling loop analysis module is configured to identify the initial charge pulses of discrete events in the loop current sequence, construct the initial charge pulse sequence, and obtain the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence. Then, the equivalent recovery time constant characterizing the recovery capability of the power distribution network is derived from the decoupling recovery limitation index. The parameter correction module is configured to construct a time-domain filtering mask and an energy recovery deduction term using the equivalent recovery time constant, and to perform dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data, so as to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network. The power protection control module is configured to take the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, calculate the battery health score through a particle swarm optimization algorithm model, and compare the battery health score with a preset judgment threshold to generate an abnormal behavior identification result.

[0008] Secondly, an abnormal behavior recognition method based on a smart door lock, applied to the aforementioned abnormal behavior recognition system based on a smart door lock, includes: During the continuous authentication failure of the smart door lock, the battery terminal voltage sequence and the loop current sequence are collected simultaneously. Identify the initial charge pulses of discrete events in the loop current sequence, construct the initial charge pulse sequence, and obtain the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence. Then, derive the equivalent recovery time constant characterizing the recovery capability of the power distribution network from the decoupling recovery limitation index. Using the equivalent recovery time constant, a time-domain filtering mask and energy recovery deduction term are constructed to perform dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data, so as to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network. Using the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, the battery health score is calculated through a particle swarm optimization algorithm model. The battery health score is then compared with a preset judgment threshold to generate an abnormal behavior identification result.

[0009] The beneficial effects of this invention include: by monitoring the transient charge recovery capability of the power supply circuit under continuous high-frequency load, the dynamic recovery time parameters of the power distribution network are derived in reverse, and dynamic interference compensation is performed on the battery voltage drop rate and cumulative energy consumption trend accordingly. This effectively eliminates false decay signals caused by limited recovery of the decoupling circuit, thus accurately distinguishing whether continuous authentication failures are due to illegal intrusion or erroneous operation caused by limited battery performance in extreme scenarios such as wireless flooding attacks. This invention not only significantly reduces the false alarm rate caused by temporary battery fatigue, but also improves the user experience and device availability under edge conditions such as low battery or battery aging by implementing a differentiated strategy response mechanism: immediate alarm for illegal intrusion and activation of the TYPE-C interface to utilize external mobile phones for authentication data collection when battery performance is limited. This ensures security while greatly improving the user experience and device availability under edge conditions such as low battery or battery aging. Attached Figure Description

[0010] Figure 1 This is a flowchart of an abnormal behavior recognition system based on a smart door lock according to the present invention; Figure 2 This is a schematic diagram of a TYPE-C interface circuit in one embodiment of the present invention; Figure 3 This is a schematic diagram of the main control circuit in one embodiment of the present invention; Figure 4 This is a schematic diagram of the output control circuit in one embodiment of the present invention. Detailed Implementation

[0011] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0012] Example 1: As Figure 1As shown, an abnormal behavior recognition system based on a smart door lock includes an electrical parameter acquisition module, a decoupling loop analysis module, a parameter correction module, and a power protection control module. The electrical parameter acquisition module is configured to simultaneously acquire the battery terminal voltage sequence and the loop current sequence during continuous authentication failures of the smart door lock. The decoupling loop analysis module is configured to identify the initial charge pulses of discrete events in the loop current sequence, construct the initial charge pulse sequence, and obtain the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence. Then, the equivalent recovery time constant characterizing the recovery capability of the power distribution network is derived from the decoupling recovery limitation index. The parameter correction module is configured to construct a time-domain filtering mask and an energy recovery deduction term using the equivalent recovery time constant, and to perform dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data, so as to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network. The power protection control module is configured to take the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, calculate the battery health score through a particle swarm optimization algorithm model, and compare the battery health score with a preset judgment threshold to generate an abnormal behavior identification result.

[0013] Preferably, the electrical parameter acquisition module is configured to simultaneously acquire the battery terminal voltage sequence and the loop current sequence during continuous authentication failures of the smart lock, including: Set sampling frequency The following conditions must be met:

[0014] According to the sampling frequency, the time variable Discretize to obtain the sampling time ,in For sampling point index; Generate the battery terminal voltage sequence respectively. and the loop current sequence :

[0015]

[0016] in, This represents an analog voltage function at the power input terminal of the smart door lock system. The analog current function representing the overall battery circuit; Furthermore, the battery terminal voltage sequence and the loop current sequence satisfy the following time synchronization constraint:

[0017] in, Indicates the same index The absolute time difference between the voltage sampling time and the current sampling time.

[0018] The sampling frequency is the signal discretization frequency set by the smart door lock's electrical parameter acquisition module when synchronously acquiring voltage and current signals. 50 kHz is preferred because the current profile of Bluetooth Low Energy (BLE) events is in the microsecond to millisecond range. A sampling frequency of 50 kHz can fully analyze the event's initial spikes and voltage transient drops, while also considering hardware power consumption and data processing pressure. A minimum acceptable frequency of 10 kHz can meet basic sampling requirements.

[0019] The battery terminal voltage sequence is a numerical sequence obtained by discretizing the analog voltage signal at the power input terminal of the smart door lock system according to a set sampling frequency. It can be obtained by connecting a high-precision voltage sensor in parallel at the input terminal of the system power management chip (PMIC) and using an analog-to-digital converter (ADC) to convert the analog signal into a digital signal.

[0020] The loop current sequence is a numerical sequence obtained by discretizing the analog current signal of the main battery loop of the smart door lock according to a set sampling frequency. It can be obtained by connecting a low-resistance sampling resistor in series in the main battery loop, detecting the voltage drop across the resistor and converting it through an analog-to-digital converter, or by directly acquiring the data using an integrated current sensor.

[0021] Time synchronization error is the absolute time difference between the voltage sampling time and the corresponding current sampling time under the same sampling index, and its value must be less than or equal to the reciprocal of the sampling frequency.

[0022] It should be noted that consecutive authentication failures include at least five consecutive authentication failures. The authentication methods may include: facial recognition, fingerprint recognition, and password recognition.

[0023] Synchronous data collection is limited to periods of continuous authentication failure for smart locks because, in this scenario, devices may experience high-density, short-duration current events due to external interference (such as Bluetooth Low Energy denial-of-service attacks). During this time, the decoupling and recovery characteristics of the power distribution network are most readily apparent, and the collected data accurately reflects the transient characteristics required for subsequent analysis. The synchronization error must be less than or equal to the reciprocal of the sampling frequency because the synchronization of voltage and current directly affects the accuracy of energy integral calculations. Excessive synchronization error can lead to deviations in subsequent energy-related parameter calculations. The requirement that the synchronization error be less than or equal to the reciprocal of the sampling frequency is based on engineering practice in transient electrical parameter measurement, ensuring that the sampled data accurately reflects the dynamic response of the power circuit. For example, when the sampling frequency is 50 kHz and the sampling period is 20 microseconds, the synchronization error must be controlled within 20 microseconds to guarantee the correspondence between voltage and current at the same time point.

[0024] The recommended sampling frequency is 50 kHz, which covers the transient characteristics of most low-power Bluetooth events in smart locks. If hardware resources are limited, the minimum should not be lower than 10 kHz. For voltage acquisition, a voltage sensor with an accuracy of not less than 0.5% (such as the ADS1115 analog-to-digital converter chip) can be used. For current acquisition, a sampling resistor with a resistance of 10 milliohms and an accuracy of 1% or an integrated current sensor of model ACS712 can be used. The specific physical point of the system power input is the power input pin of the power management chip (PMIC). When wiring, the distance between the sensor and the acquisition point should be minimized to reduce line interference. An anti-aliasing low-pass filter should be configured during sampling, with the cutoff frequency set to 0.45 times the sampling frequency. For example, when sampling at 50 kHz, the filter cutoff frequency is 22.5 kHz to avoid high-frequency noise affecting the sampling accuracy.

[0025] Preferably, the decoupling loop analysis module is configured to identify the initial charge pulses of discrete events in the loop current sequence, and to invert the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence, including: Determine the loop current sequence The starting point of the event :

[0026] in The ninety percentile of the loop current sequence is used as the trigger threshold. Filter to meet and The time index is used to exclude items with time intervals shorter than the minimum interval. The index is used to obtain the set of event origins. ; Calculate the initial charge pulse corresponding to the start of each event. :

[0027]

[0028] in The 10 percentile of the loop current sequence is used as the standby baseline current. This corresponds to the number of sampling points within the preset extremely short integration window; The sampling frequency; Based on the initial charge pulse sequence Calculate the decoupling recovery limitation index :

[0029]

[0030]

[0031] in It is a first-order difference sequence. To prevent the division by zero regularization term, It is a sequence of difference ratios. This indicates the operation of taking the median.

[0032] The trigger threshold is a discrete event identification standard determined based on the high quantile value of the loop current sequence. It is preferably the 90th percentile of the loop current sequence because during consecutive authentication failures, the event current is typically in the high quantile range of the current distribution. This quantile can stably separate the event from the background current and accommodate the differences in current characteristics of different door locks.

[0033] The minimum event interval constraint is a time interval threshold set to avoid the repeated marking of multiple peaks of the same discrete event. It is preferably 5 milliseconds because the phased nature of Bluetooth Low Energy events means that the duration of a single event is usually no more than 5 milliseconds, and this interval can effectively eliminate the starting point of repeated marking.

[0034] The event start index is the sequence index corresponding to the moment in the loop current sequence when the amplitude exceeds the trigger threshold and meets the minimum time interval constraint.

[0035] The standby baseline current is a reference current for the device in standby mode, determined based on the low quantile value of the loop current sequence. It is preferably a tenth of a percentile of the loop current sequence, because the standby current is typically in the low quantile range of the current distribution. This quantile can avoid interference from event currents on the baseline, ensuring baseline stability.

[0036] The preset number of sampling points for the ultra-short integration window is the number of sampling points corresponding to the fixed integration window used to extract the initial charge pulse. Preferably, it is 25 sampling points, because the duration of the initial spike of a Bluetooth Low Energy event is approximately 0.5 milliseconds. When the sampling frequency is 50 kHz, 25 sampling points can completely cover the charge characteristics of the initial spike.

[0037] The initial charge pulse is the amount of charge obtained by summing and integrating the difference between the loop current sample and the standby baseline current within a preset extremely short integration window after the event starts, reflecting the charge demand at the beginning of the event.

[0038] A first-order difference sequence is a sequence composed of the differences between two adjacent pulse values ​​in the initial charge pulse sequence, used to characterize the changing trend of charge pulses.

[0039] The division-to-zero regularization term is a minimal constant introduced to prevent the occurrence of zero values ​​in the first-order difference sequence, which could lead to divergence in the ratio calculation. It is preferably 10 to the power of negative 12 coulombs, a value much smaller than the differential change of the actual charge pulse, thus not affecting the accuracy of the ratio calculation and effectively avoiding division-to-zero errors.

[0040] The difference ratio is the ratio of two adjacent terms in a first-order difference sequence, obtained after removing zero regularization terms, and is used to reflect the stability of the charge pulse change trend.

[0041] The decoupling recovery limitation index is the median of all differential ratios, used to quantify the recovery capability of decoupling loops in power distribution networks.

[0042] The high quantile (90%) is used as the trigger threshold, and the low quantile (10%) as the standby baseline current, instead of a fixed threshold. This is because different smart locks have different battery types and load characteristics, making fixed thresholds less adaptable. Quantile thresholds, on the other hand, can adapt to the current distribution characteristics of different devices, improving the universality of event recognition. The initial charge pulse is extracted by accumulating and integrating the current through a preset extremely short integration window because the initial stage of a Bluetooth Low Energy event has a characteristic spike related to decoupling capacitor charging. This spike has a short duration, and the extremely short window integration can accurately capture its charge characteristics. A zero-prevention regularization term is introduced, and the median of the difference ratio is taken because the first-order difference sequence may have zero values ​​or abnormal extreme values ​​due to noise. The regularization term avoids calculation divergence, and the median suppresses outlier interference, ensuring the stability of the decoupling recovery constraint index. This processing logic is customized for the current characteristics of high-frequency events in door locks and is not a general data processing method. For example, when the 90th percentile of the loop current sequence is 200 microamps, setting the trigger threshold to 200 microamps can effectively identify current pulses caused by low-power Bluetooth connection events.

[0043] The preset duration of the ultra-short integration window is preferentially set to 0.5 milliseconds. The corresponding number of sampling points is calculated based on the sampling frequency. For example, if the sampling frequency is 50 kHz, the number of sampling points is 25; if the sampling frequency is 10 kHz, the number of sampling points is 5, ensuring that the integration window always covers the initial spike of the Bluetooth Low Energy event. The specific value of the minimum event interval constraint can be adjusted within the range of 3 to 7 milliseconds, determined according to the Bluetooth Low Energy chip model used in the door lock. When the chip event cycle is long, the interval can be appropriately increased. The specific value of the zero-removal regularization term can be selected within the range of 10 to the power of -12 coulombs to 10 to the power of -10 coulombs. The value must be much smaller than the differential change of the initial charge pulse in actual applications to avoid substantial impact on the comparison value calculation. When identifying the event start point, the indices that satisfy the condition that the current of the previous sampling point is less than the trigger threshold and the current of the current sampling point is greater than or equal to the trigger threshold must be screened first. Then, duplicate indices are eliminated according to the minimum event interval constraint. The elimination logic is: if the time difference between two adjacent candidate indices is less than the minimum event interval, the index that appears first is retained.

[0044] Preferably, the equivalent recovery time constant characterizing the recovery capability of the power distribution network is derived from the decoupling recovery-limited index, including: Calculate the average event interval :

[0045] in, For the first The moment when the event begins. This indicates the median operation; Based on the decoupling recovery limitation index Calculate the equivalent recovery time constant. :

[0046]

[0047] in, Represents the natural logarithm operation. The original recovery time constant, and These are the preset minimum time constant threshold and maximum time constant threshold, respectively. and These represent the operations of taking the minimum value and taking the maximum value, respectively.

[0048] An event interval sequence is a numerical sequence consisting of the time difference between two adjacent event start points in an event start sequence, used to characterize the occurrence density of discrete events.

[0049] The average event interval is the median of the event interval sequence, used to characterize the average occurrence period of discrete events during a series of authentication failures.

[0050] The original recovery time constant is a parameter obtained by dividing the average event interval by the natural logarithm of the decoupling recovery constraint index and taking the opposite number. It reflects the initial recovery time of the power distribution network.

[0051] The minimum time constant threshold is a preset lower limit value to ensure that the physical meaning of the equivalent recovery time constant is reasonable. It is preferably 0.1 milliseconds, because the typical recovery time of the decoupling capacitor in the power distribution network of the smart door lock will not be shorter than this value, avoiding abnormally small values ​​that may cause subsequent interference compensation failure.

[0052] The maximum time constant threshold is a preset upper limit value to ensure that the equivalent recovery time constant has a reasonable meaning. It is preferably 5 seconds, because the event interval during continuous authentication failures of the door lock is usually in the range of milliseconds to seconds, and a recovery time constant exceeding this value has no practical engineering significance.

[0053] The equivalent recovery time constant is a parameter obtained by truncating the original recovery time constant between the minimum time constant threshold and the maximum time constant threshold. It is used to quantify the actual recovery capability of the power distribution network.

[0054] The equivalent recovery time constant is derived by considering the decoupling recovery limitation index and the average event interval, based on the first-order RC recovery characteristics of the decoupling capacitor in the power distribution network. The decoupling capacitor recovers through charging in the power distribution network during the event interval, and its recovery process follows an exponential decay law. The natural logarithm is the inverse operation of this exponential relationship, thus establishing the correlation between event density and recovery capability. The median is used to calculate the average event interval because the event interval may fluctuate abnormally due to interference. The median can suppress extreme interference, ensuring that the average event interval accurately reflects the stable period of event occurrence. For example, when the decoupling recovery limitation index is 0.92 and the average event interval is 25 milliseconds, the original recovery time constant is calculated to be approximately 296 milliseconds. This value is within a reasonable range and accurately reflects the recovery characteristics of the decoupling loop.

[0055] The minimum time constant threshold is fixed at 0.1 milliseconds, and the maximum time constant threshold is fixed at 5 seconds. This range covers the decoupling recovery time interval of mainstream smart door lock power distribution networks. The sample size of the event interval sequence must include at least 5 time differences between adjacent events. If the number of events in the event starting sequence is less than 6 (i.e., the event interval is less than 5), more events must be generated before calculating the average event interval to ensure the reliability of the median statistics. The calculation of the original recovery time constant must use the natural logarithm function. In engineering implementation, the natural logarithm interface in the standard mathematical library can be called. The calculation precision must retain at least 6 decimal places to avoid numerical truncation causing deviation of the equivalent recovery time constant. The truncation processing logic is as follows: if the original recovery time constant is less than 0.1 milliseconds, the equivalent recovery time constant is 0.1 milliseconds; if it is greater than 5 seconds, it is 5 seconds; if it is within the range, the original value remains unchanged.

[0056] Preferably, the parameter correction module is configured to construct a time-domain filtering mask and an energy recovery deduction term using the equivalent recovery time constant, and to perform dynamic interference compensation on the voltage recovery hysteresis rate calculated based on the battery terminal voltage sequence and the cumulative energy consumption slope calculated based on the voltage-current product, respectively, to remove the dynamic interference introduced by the limited recovery of the power distribution network, including: Construct the temporal filtering mask :

[0057] in, To restore the time index of the fragment, The sampling frequency is... The equivalent recovery time constant is... Calculate the voltage recovery hysteresis rate :

[0058]

[0059]

[0060]

[0061] in, For the first Index of voltage valley times for discrete events To recover the voltage residual, This represents the effective recovery amount after masking. The hysteresis rate for a single event. To prevent the logarithm from being a singular, extremely small constant, To restore the window length; Calculate the cumulative energy consumption slope :

[0062]

[0063]

[0064]

[0065]

[0066] in, For the first The original energy of a discrete event. For the energy replenishment deduction item, To remove bias energy, For the cumulative energy consumption sequence, The initial charge pulse, The average event interval is... The equivalent capacitance in hardware. The loss coefficient is... The length of the energy window. The total number of events, and These are the event sequence number and the mean of the accumulated energy consumption sequence, respectively.

[0067] The time-domain filtering mask is an exponential decay function constructed based on the equivalent recovery time constant, used to remove the early recovery component dominated by the power distribution network in the voltage recovery residual.

[0068] The voltage trough moment index is the sequence index of the voltage sequence corresponding to each discrete event, indicating the position where the voltage drops to the trough.

[0069] Voltage recovery residual is the difference between the voltage at each sampling point after the voltage trough and the trough voltage, reflecting the original change during the voltage recovery process.

[0070] The effective recovery value is the value obtained after the voltage recovery residual is weighted by a time-domain filtering mask, which removes dynamic interference from the power distribution network.

[0071] The logarithmic singularity prevention constant is a fixed constant introduced to avoid calculation anomalies caused by zero or minimum values ​​in logarithmic operations. It is preferably 10 to the power of negative 6 volts, a value far smaller than the actual fluctuation range of the door lock voltage recovery residual, thus not affecting the calculation and effectively avoiding the logarithmic singularity problem.

[0072] The recovery window length is the number of sampling points corresponding to a fixed time window used to extract the recovery segment after truncation of the voltage valley. A preferred number of sampling points is 20 milliseconds, because the main stage of the smart door lock voltage recovery process is usually completed within 20 milliseconds, and this window can fully cover the recovery characteristics. For example, at a sampling frequency of 50 kHz, this corresponds to 1000 sampling points.

[0073] The voltage recovery hysteresis rate is the median of the logarithmic slope of the effective recovery amount of all discrete events, and is used to characterize the voltage recovery characteristics of the battery itself.

[0074] The preset energy window sampling point number is the number of sampling points corresponding to a fixed time window used to calculate the raw energy of a single discrete event. A sampling point number corresponding to 30 milliseconds is preferred because the complete energy consumption process of a Bluetooth Low Energy event typically does not exceed 30 milliseconds, and this window can comprehensively capture the energy characteristics of the event.

[0075] The raw energy is the cumulative integral of the product of voltage and current for each discrete event within a preset energy window, reflecting the total energy consumption of the event.

[0076] The hardware equivalent capacitance is an equivalent parameter characterizing the combined energy storage capacity of the decoupling capacitor and the power input capacitor in the power distribution network of a smart door lock. Preferred values ​​are determined through factory calibration, based on the hardware design scheme of the door lock power network, and are typically in the microfarad to millifarad range.

[0077] The loss factor is an equivalent proportional coefficient that comprehensively considers the equivalent series resistance loss, DC-DC conversion efficiency, and wiring loss in the power distribution network. The preferred value range is 0.5 to 3, based on the loss characteristics of mainstream smart door lock power networks as specified in the factory calibration.

[0078] The recovery limitation factor is a parameter calculated based on the average event interval and the equivalent recovery time constant, used to quantify the adequacy of the charging recovery of the decoupling capacitor.

[0079] The energy recovery deduction is used to deduct the extra energy consumed by decoupling capacitors in the power distribution network for power recovery.

[0080] The debiased energy is the value obtained by subtracting the energy replenishment deduction from the original energy, reflecting the actual energy consumption of the battery and the load.

[0081] The accumulated energy consumption sequence is a sequence obtained by accumulating the debiased energy of all discrete events in the order in which the events occur.

[0082] The total number of events is the total number of discrete events identified during a period of consecutive authentication failures.

[0083] The event index mean is the arithmetic mean of all discrete event indexes.

[0084] The cumulative energy consumption sequence mean is the arithmetic mean of all values ​​in the cumulative energy consumption sequence.

[0085] The cumulative energy consumption slope is the slope of the linear regression of the cumulative energy consumption sequence with respect to the event sequence number, used to characterize the trend of energy consumption changes with the accumulation of events.

[0086] An exponentially decaying time-domain filtering mask is constructed based on the equivalent recovery time constant. The core logic is that the recovery process of the decoupling capacitor in the power distribution network follows the exponential decay law of a first-order RC circuit, and the equivalent recovery time constant directly reflects this recovery characteristic. The mask's weight decreases with increasing recovery time, accurately eliminating early components dominated by decoupling recovery while retaining effective information about battery polarization recovery. The energy recovery deduction term integrates the square of the initial charge pulse, the hardware equivalent capacitance, the loss coefficient, and the recovery limiting factor. This is because the energy storage of the decoupling capacitor is proportional to the square of the charge, the recovery limiting factor reflects the impact of the energy quantification event interval on recovery, and the loss coefficient compensates for energy loss in the actual circuit. This construct establishes a correlation between charge characteristics, hardware parameters, and energy loss, which is not a general energy consumption calculation method. The hysteresis rate is calculated using a logarithmic slope and the median is taken because the voltage recovery process is approximately exponential, and logarithmic transformation results in higher linearity. The median can suppress interference from abnormal events, ensuring the stability of the performance indicators. For example, when the equivalent recovery time constant is 296 milliseconds, the weight of the temporal filtering mask is close to 1 in the early stage of recovery, which can effectively suppress the decoupling-dominated recovery component.

[0087] The minimum constant for preventing logarithmic singularities is fixed at 10 to the power of -6 volts, a value applicable to voltage measurement scenarios for all mainstream smart locks. The recovery window length is fixed at 20 milliseconds, with the corresponding number of sampling points calculated based on the sampling frequency; for example, 200 sampling points for a sampling frequency of 10 kHz. The preset energy window duration is fixed at 30 milliseconds, with the corresponding number of sampling points being the sampling frequency multiplied by 30 milliseconds; for example, 1500 sampling points for 50 kHz sampling. The calibration method for the hardware equivalent capacitance is as follows: apply a pulse current of fixed amplitude and duration to the test fixture, measure the transient voltage drop at the power input terminal, and calculate the capacitance using the ratio of charge to voltage drop. The calibration method for the loss coefficient is as follows: under the same fixed event, measure the ratio of the energy change at the battery terminal before and after the event to the theoretical energy stored in the decoupling capacitor, and trim it to a range of 0.5 to 3 as the final value. The logic for locating the voltage valley moment index is as follows: The earliest sampling point corresponding to the 5th percentile of the voltage within a 2-millisecond window after the event start point is selected to avoid noise interference at a single extreme point. The linear regression slope is calculated using the least squares method, with a calculation precision retaining at least 6 decimal places.

[0088] Preferably, the power protection control module is configured to use the voltage recovery hysteresis rate after the dynamic interference compensation processing and the cumulative energy consumption slope as input parameters to calculate the battery health score through a particle swarm optimization algorithm model, including: Calculate the discrimination score based on the optimal parameters of the particle swarm optimization algorithm model. :

[0089] in, The voltage recovery hysteresis rate, The cumulative energy consumption slope is... and For optimal weight parameters, and The optimal threshold parameter is... and The preset temperature coefficient, It is a sigmoid function; Based on the equivalent recovery time constant Calculate the monotonic modulation factor :

[0090] in, The preset modulation constant, It represents exponential operations with the natural constant as the base; Calculate the battery health score :

[0091]

[0092] in, and These represent the operations of taking the minimum value and taking the maximum value, respectively.

[0093] The voltage recovery hysteresis rate is the median logarithmic slope of the battery body voltage recovery characteristics obtained after the parameter correction module removes the dynamic interference of the power distribution network.

[0094] The cumulative energy consumption slope is a linear regression slope that represents the trend of energy consumption changes with the cumulative events, obtained after the parameter correction module removes the dynamic interference of the power distribution network.

[0095] The optimal weighting parameter for the voltage recovery hysteresis rate is the output of the particle swarm optimization algorithm model and is used to adjust the contribution ratio of the voltage recovery hysteresis rate in the discrimination score.

[0096] The optimal weighting parameter of the cumulative energy consumption slope is the output of the particle swarm optimization algorithm model. It is used to adjust the contribution ratio of the cumulative energy consumption slope in the discrimination score, and its sum with the optimal weighting parameter of the voltage recovery hysteresis rate is 1.

[0097] The optimal threshold parameter for voltage recovery hysteresis rate is the benchmark parameter output by the particle swarm optimization algorithm model and is used for voltage recovery hysteresis rate normalization processing.

[0098] The optimal threshold parameter for the cumulative energy consumption slope is the benchmark parameter output by the particle swarm optimization algorithm model and is used for the normalization of the cumulative energy consumption slope.

[0099] The temperature coefficient of the voltage recovery hysteresis rate is a preset parameter used to compensate for the effect of temperature changes on the voltage recovery hysteresis rate. It is preferably the interquartile range of the voltage recovery hysteresis rate in historical samples, because the interquartile range can stably reflect the degree of data dispersion and adapt to parameter fluctuations at different temperatures.

[0100] The temperature coefficient of the cumulative energy consumption slope is a preset parameter used to compensate for the effect of temperature changes on the cumulative energy consumption slope. It is preferably the interquartile range of the cumulative energy consumption slope in historical samples, thus aligning with the temperature coefficient of the voltage recovery hysteresis rate.

[0101] The sigmoid function is a smooth function used to map the normalized voltage recovery hysteresis rate and cumulative energy consumption slope to the interval between 0 and 1, thereby achieving nonlinear discrimination.

[0102] The discrimination score is a value obtained by weighting and summing the normalized voltage recovery hysteresis rate and cumulative energy consumption slope using an S-shaped function and the optimal weight parameters. It reflects the degree of battery abnormality.

[0103] The modulation constant is a preset parameter used to adjust the influence of the equivalent recovery time constant on the health score. Its value is preferentially determined by calibration using historical normal samples, ensuring that the monotonic modulation factor for normal samples is 0.9, thus avoiding excessive attenuation of the normal state score.

[0104] The monotonic modulation factor is an exponential decay term calculated based on the equivalent recovery time constant, used to quantify the impact of the limited recovery of the power distribution network on the battery health score.

[0105] The original battery health score is the difference between the numerical value and the discrimination score, multiplied by one hundred, and then multiplied by the monotonic modulation factor to obtain a score that has not been limited by range.

[0106] The battery health score is the final score obtained by limiting the original battery health score to the range of zero to one hundred, and it is used to map the graded power protection strategy.

[0107] The optimal weights and thresholds obtained from particle swarm optimization are combined with a sigmoid function to process the voltage recovery hysteresis rate and cumulative energy consumption slope after bias removal. Since battery health is not a binary determination, the sigmoid function enables a smooth transition from normal to abnormal, while the optimal weights and thresholds adapt to the hardware characteristics and data distribution of different door locks. This combination improves the adaptability of the judgment. A monotonic modulation factor based on the equivalent recovery time constant is introduced because the recovery capability of the power distribution network directly affects power supply stability. Even if the battery itself is in good condition, limited recovery increases the risk to battery life. The monotonic modulation factor transforms loop characteristics into score decay, allowing the score to reflect both the battery itself and the loop status, spanning battery technology and power integrity domains. For example, when the equivalent recovery time constant is 296 milliseconds and the modulation constant is calibrated to 3.47 seconds per second, the monotonic modulation factor is approximately 0.9, with only a slight decay in the normal score; if the recovery time constant increases to 1 second, the modulation factor is approximately 0.03, and the score drops significantly, reflecting loop fragility.

[0108] The temperature coefficient of the voltage recovery hysteresis rate is specifically taken as the interquartile range of the voltage recovery hysteresis rate in the historical samples, calculated by subtracting the first quartile from the third quartile; the temperature coefficient of the cumulative energy consumption slope is calculated in the same way. The calibration steps for the modulation constant are as follows: select the median of the equivalent recovery time constant of normal samples in the historical data, set the monotonic modulation factor corresponding to this median to 0.9, and inversely deduce the modulation constant through exponential operations. For example, when the median of normal samples is 296 milliseconds, the modulation constant is 3.47 per second. The key parameters of the particle swarm optimization algorithm are set as follows: number of particles 30, number of iterations 60, inertia weight 0.7, cognitive coefficient and social coefficient both 1.7, and the upper limit of velocity is 0.2 times the range of each parameter. The historical sample set must contain at least 500 sets of continuous authentication failure window data under different door locks, different battery states, and different ambient temperatures. The labeling rule is: if the continuous failure window disappears within 24 hours after replacing the battery, it is labeled as non-abnormal; otherwise, it is labeled as abnormal. The calculation precision of the S-shaped function is retained to 6 decimal places.

[0109] Preferably, the battery health score is compared with a preset judgment threshold to generate an abnormal behavior identification result, including: Set the preset judgment threshold as ; If the battery health score Conditions met:

[0110] The continuous authentication failures are then determined to be abnormal behavior caused by unauthorized intrusion. If the battery health score Conditions met:

[0111] The continuous authentication failures are then determined to be abnormal behavior caused by limited battery performance.

[0112] The preset threshold is a critical score used to distinguish whether the battery is healthy. This threshold represents the minimum health standard required for the battery to maintain normal operation after eliminating interference from decoupling capacitors that limit recovery due to high-frequency use. The preset threshold is typically set to 60 points. A battery health score below this indicates a significant increase in battery internal resistance, which may cause instability in the smart lock's electrical components, leading to continuous authentication failures. A battery health score above this indicates stable electrical component operation; in this case, continuous authentication failures may indicate unauthorized intrusion, such as a stranger attempting to unlock the smart lock.

[0113] Specifically, the policy generation module is configured to execute the following logic: When the continuous authentication failures are determined to be abnormal behavior caused by unauthorized intrusion (i.e., battery health score), When the value exceeds the preset threshold, the system determines that the battery power supply is normal, but it has encountered a high-frequency external attack or malicious trial and error.

[0114] Triggering mechanism: The policy generation module receives a determination signal indicating an illegal intrusion.

[0115] Action executed: The module immediately generates an alarm command and activates the alarm module it communicates with.

[0116] Specific handling procedures (proactive defense strategies in illegal intrusion scenarios): Local deterrence: Control the built-in buzzer of the door lock to emit a high-decibel alarm sound, or control the LED warning lights on the front panel to flash red and blue at high frequency, physically deterring intruders. Remote notification: Send an abnormal intrusion alarm push to the user's bound mobile terminal or cloud server through the smart door lock's wireless communication unit (such as Wi-Fi or ZigBee module), informing the user that the door lock is currently under attack. System locking: While executing the alarm, the policy generation module can issue a command to temporarily cut off the power supply circuit of the biometric module (fingerprint, face), putting the door lock into a forced sleep lock state (e.g., locked for 3 minutes), preventing attackers from continuing to attempt to crack the system.

[0117] Specific processing flow (emergency assistance strategy for scenarios with limited battery performance): When the continuous authentication failure is determined to be an abnormal behavior caused by limited battery performance (i.e., battery health score) When the value is lower than or equal to the preset threshold, the system determines that the damage is not due to malicious intent, but rather to an increase in battery internal resistance or depletion of power, which prevents the system from supporting the large instantaneous current required for fingerprint collection, facial recognition, or motor drive, thus causing authentication failure.

[0118] Therefore, the policy generation module will execute the following special wired-assisted authentication process: The strategy generation module activates the TYPE-C interface at the bottom of the smart lock's front panel. In normal standby mode, this interface is off, but under this abnormal trigger, it is configured to OTG mode with data transmission and reverse power handshake capabilities. The voice module connected to it is activated, playing a specific prompt, such as: "Battery aging detected, authentication cannot be completed. Please insert your phone into the emergency interface to assist in unlocking." This indicates to the initiator of the abnormal behavior that the lock is not damaged, but requires external assistance. When the user (initiator) connects their phone to the lock's TYPE-C interface via a data cable, the lock identifies the device through a handshake protocol. At this point, the data connection not only obtains power (traditional emergency charging) but also establishes a data channel. Specifically, this includes: computing power / energy consumption transfer. Since the lock's internal battery cannot support the high-power authentication process, the strategy generation module transfers some of the data acquisition and processing tasks required for authentication to external input.

[0119] The strategy generation module executes as follows: Bypass power supply for data acquisition: The door lock obtains a stable 5V power supply from the mobile phone through the TYPE-C interface, directly bypassing the internal aging battery, and independently powering the fingerprint sensor or face camera, thereby completing a high-quality biometric data acquisition and comparison without replacing the battery.

[0120] Authorization signaling interaction: If the door lock sensor is not working, the system can request the mobile APP to authenticate the identity (such as entering a password or verifying biometrics on the mobile phone). After the mobile phone is verified, it sends an encrypted unlocking authorization command to the door lock through the TYPE-C interface.

[0121] Example 2: An abnormal behavior recognition method based on smart door locks, applied to an abnormal behavior recognition system based on smart door locks, including: During the continuous authentication failure of the smart door lock, the battery terminal voltage sequence and the loop current sequence are collected simultaneously. Identify the initial charge pulses of discrete events in the loop current sequence, construct the initial charge pulse sequence, and obtain the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence. Then, derive the equivalent recovery time constant characterizing the recovery capability of the power distribution network from the decoupling recovery limitation index. Using the equivalent recovery time constant, a time-domain filtering mask and energy recovery deduction term are constructed to perform dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data, so as to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network. Using the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, the battery health score is calculated through a particle swarm optimization algorithm model. The battery health score is then compared with a preset judgment threshold to generate an abnormal behavior identification result.

[0122] Example 3: Based on the above examples, to further improve the user experience and unlocking success rate in scenarios where battery performance is severely limited, this invention also provides an enhanced emergency power supply and execution guarantee strategy. After the system determines that continuous authentication failures are caused by limited battery performance, this strategy not only activates the TYPE-C interface for auxiliary authentication data interaction, but also uses the same interface to charge the built-in supercapacitor to provide the transient high current required by the unlocking motor, ensuring that the door lock can still reliably perform the unlocking action even when the battery's power supply is insufficient.

[0123] like Figure 2 As shown in the diagram (TYPE-C interface and enhanced emergency power supply circuit), this circuit includes a 6-pin TYPE-C female connector, resistors RU and RU1, and a transient voltage suppressor diode TVS1. Resistors RU and RU1 ensure a fixed 5V output from external power banks, mobile phones, or other power supply devices, guaranteeing the safe and stable operation of subsequent circuits. TVS1 absorbs occasional high-voltage spikes from external power supplies, preventing damage to downstream circuits.

[0124] like Figure 3 The diagram shown is a schematic of a supercapacitor charging circuit (i.e., the main control circuit) according to one embodiment of the present invention. This circuit includes an input power supply positive terminal C+ and a negative terminal C-, and... Figure 2 The circuit is connected to the output terminal of the TYPE-C interface. Specifically, it includes: a charging IC U1, an inductor L1, a Schottky diode D1, a MOSFET Q2, multiple resistors (R16, RCS, R2, R1, R4, R8, etc.), capacitors (C3, C4, CO), LED indicators LED1 and LED2, and three supercapacitors P1, P2, and P3 connected in series. Its working principle is as follows: an external 5V power supply is input through C+ and C-, controlled by the charging IC U1, and boosted to approximately 8.4V through inductor L1 to charge the series-connected supercapacitors. Each supercapacitor P1, P2, and P3 has a nominal voltage of 2.7V, and the total voltage after series connection is approximately 8.1V, matching the system power supply voltage of the smart door lock. During charging, LED2 illuminates to indicate charging is in progress; once fully charged, LED1 illuminates or turns off. TVS2 is used to absorb instantaneous high voltage, protecting the subsequent door lock circuitry.

[0125] like Figure 4The diagram shown is a schematic of the output control circuit according to one embodiment of the present invention. The circuit includes a microcontroller U2, resistors R3, R5, R6, R7, R9, and R10, capacitors C1 and C2, a MOSFET Q1, and a diode D2. The microcontroller U2 is connected to the charging status indicator terminal of the charging IC U1 via resistor R3 to detect whether the supercapacitor is fully charged; the real-time voltage of the supercapacitor is detected through a voltage divider network composed of resistors R5, R6, and C2; the output terminal of the microcontroller U2 (e.g., pin 3) is connected to the gate of the MOSFET Q1 via resistor R9 to control the conduction and cutoff of Q1; the source and drain of Q1 are connected in series between the supercapacitor group (the positive terminals of P1, P2, and P3) and the power input terminal of the smart door lock system; and diode D2 prevents external voltage backflow.

[0126] The working logic of this embodiment is as follows: When the battery health score calculated by the power protection control module based on the particle swarm optimization algorithm is lower than or equal to a preset threshold, the system determines that continuous authentication failures are caused by limited battery performance. At this time, the strategy generation module generates an activation signal and performs the following operations: Activate TYPE-C interface and voice prompt: Wake up Figure 2 The TYPE-C interface circuit shown is set to OTG mode, allowing it to receive external power and perform data communication. At the same time, the voice module is activated, playing prompts to guide the user to connect an external power source (such as a mobile phone or power bank) to the TYPE-C interface of the door lock via a data cable.

[0127] Supercapacitor charging: When the user connects an external power source, the external 5V power supply... Figure 2 Circuit input to Figure 3 In the charging circuit, charging IC U1 starts working, charging the supercapacitor group P1, P2, and P3. Microcontroller U2 monitors the supercapacitor voltage in real time through a voltage divider network and obtains the charging progress information through the charging status indicator.

[0128] Full charge detection and automatic discharge: When the microcontroller U2 detects that the supercapacitor voltage has reached a preset threshold (e.g., 8V) and the charging IC sends a full charge signal, it delays for a certain period of time (e.g., 2 seconds) to ensure stability. Then, it controls the MOSFET Q1 to turn on, connecting the supercapacitor bank to the power input terminal of the door lock system. Due to its low internal resistance, the supercapacitor can instantly release a large current to meet the transient power requirements of the unlocking motor during startup, thereby driving the motor to complete the unlocking action.

[0129] Safety protection: After unlocking, the microcontroller U2 can control Q1 to turn off to prevent the supercapacitor from continuously discharging; TVS1, TVS2 and D2 and other components provide overvoltage, reverse connection and other protection to ensure circuit safety.

[0130] Through the aforementioned enhanced emergency power supply strategy, this invention, based on accurate identification of limited battery performance, not only achieves auxiliary authentication data interaction, but also utilizes the same TYPE-C interface to complete external power supply, supercapacitor energy storage, and transient high current output. This fundamentally solves the problem of "having power but being unable to unlock" caused by battery aging, increased internal resistance, or depletion of power, significantly improving the usability and user experience of smart door locks under extreme conditions.

[0131] It should be noted that, Figure 2 , Figure 3 , Figure 4 The circuit shown is only an exemplary implementation. Those skilled in the art can adjust the component parameters or circuit structure according to actual needs, such as changing the number of supercapacitors in series to adapt to different voltage requirements, or selecting different models of charging ICs, microcontrollers, etc., all of which fall within the protection scope of this invention.

[0132] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. An abnormal behavior recognition system based on a smart door lock, characterized in that, It includes an electrical parameter acquisition module, a decoupling loop analysis module, a parameter correction module, and a power protection and control module; The electrical parameter acquisition module is configured to simultaneously acquire the battery terminal voltage sequence and the loop current sequence during continuous authentication failures of the smart door lock. The decoupling loop analysis module is configured to identify the initial charge pulses of discrete events in the loop current sequence, construct the initial charge pulse sequence, and obtain the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence. Then, the equivalent recovery time constant characterizing the recovery capability of the power distribution network is derived from the decoupling recovery limitation index. The parameter correction module is configured to construct a time-domain filtering mask and an energy recovery deduction term using the equivalent recovery time constant, and to perform dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data, so as to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network. The power protection control module is configured to take the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, calculate the battery health score through a particle swarm optimization algorithm model, and compare the battery health score with a preset judgment threshold to generate an abnormal behavior identification result.

2. The abnormal behavior recognition system based on a smart door lock according to claim 1, characterized in that, During periods of continuous authentication failure of the smart lock, the electrical parameter acquisition module synchronously acquires the battery terminal voltage sequence and the loop current sequence, including: Set the sampling frequency to be greater than or equal to 10,000 Hz; The analog voltage signal at the system power input terminal of the smart door lock and the analog current signal in the battery main circuit are converted into discrete battery terminal voltage sequences and circuit current sequences respectively according to the sampling frequency. The time synchronization error between the voltage sampling time in the battery terminal voltage sequence and the corresponding current sampling time in the loop current sequence is controlled to be less than or equal to one sampling period, which is the reciprocal of the sampling frequency.

3. The abnormal behavior recognition system based on a smart door lock according to claim 2, characterized in that, The decoupling loop analysis module identifies the initial charge pulses of discrete events in the loop current sequence and inverts the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence, including: The high quantile value of the loop current sequence is calculated as the trigger threshold, and the moment when the amplitude of the loop current sequence exceeds the trigger threshold and satisfies the minimum time interval constraint is marked as the event start point of the discrete event, forming an event start point sequence; The low quantile value of the loop current sequence is calculated as the standby baseline current. The difference between the loop current sample and the standby baseline current within a preset very short integration window after each event start is accumulated and integrated to obtain the corresponding starting charge pulse. Construct a first-order difference sequence of the initial charge pulse sequence, and calculate the ratio of two adjacent terms in the first-order difference sequence, wherein a zero-prevention regularization term is introduced when calculating the ratio; Calculate the median of all the ratios and use this median as the metric for limited decoupling recovery.

4. The abnormal behavior recognition system based on a smart door lock according to claim 3, characterized in that, The decoupling loop analysis module derives an equivalent recovery time constant characterizing the recovery capability of the power distribution network from the decoupling recovery limitation index, including: Calculate the time difference between two adjacent event starting points in the event starting point sequence to obtain the event interval sequence, and calculate the median of the event interval sequence as the average event interval; Divide the average event interval by the natural logarithm of the decoupling recovery constraint index and take the negative of the result to obtain the original recovery time constant; Determine whether the original recovery time constant is between the preset minimum time constant threshold and the maximum time constant threshold. If it exceeds the range, truncate its value to the corresponding threshold boundary to obtain the equivalent recovery time constant.

5. The abnormal behavior recognition system based on a smart door lock according to claim 4, characterized in that, The parameter correction module utilizes the equivalent recovery time constant to construct a time-domain filtering mask and an energy recovery deduction term, and performs dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network, including: An exponential decay function based on the equivalent recovery time constant is constructed as the time-domain filtering mask; the voltage valley moment of each discrete event is located in the battery terminal voltage sequence, the voltage recovery segment after the voltage valley moment is extracted to calculate the voltage recovery residual, the voltage recovery residual is weighted using the time-domain filtering mask to remove early recovery components, the effective recovery amount is obtained, the logarithmic slope of the effective recovery amount is calculated and the median is taken as the voltage recovery hysteresis rate; The integral of the voltage-current product of each discrete event within a preset energy window is calculated as the raw energy; The energy recovery deduction term is constructed using the square of the initial charge pulse, the preset hardware equivalent capacitance, the preset loss coefficient, and the recovery limiting factor calculated based on the average event interval and the equivalent recovery time constant. The original energy is subtracted from the energy replenishment deduction term to obtain the debiased energy. The cumulative sequence of the debiased energy is calculated, and the linear regression slope of the cumulative sequence with respect to the event sequence number is calculated as the cumulative energy consumption slope.

6. The abnormal behavior recognition system based on a smart door lock according to claim 5, characterized in that, The power protection control module uses the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, and calculates the battery health score using a particle swarm optimization algorithm model, including: Obtain the optimal weight parameters and optimal threshold parameters output by the particle swarm optimization algorithm model; Using the optimal threshold parameter and the preset temperature coefficient, the voltage recovery hysteresis rate and the cumulative energy consumption slope are normalized and substituted into the S-shaped function. The output of the S-shaped function is then weighted and summed using the optimal weight parameter to obtain the discrimination score. Using the equivalent recovery time constant and the preset modulation constant, the exponential attenuation term with the natural constant as the base is calculated to obtain the monotonic modulation factor; Calculate the difference between the numerical value and the discrimination score, multiply the difference by the numerical value of one hundred, and then multiply the monotonic modulation factor to obtain the target product. Limit the target product to the range of zero to one hundred to obtain the battery health score.

7. The abnormal behavior recognition system based on a smart door lock according to claim 6, characterized in that, The battery health score is compared with a preset judgment threshold to generate an abnormal behavior identification result, including: If the battery health score is higher than a preset threshold, the continuous authentication failure is determined to be an abnormal behavior caused by illegal intrusion; if the battery health score is lower than or equal to the preset threshold, the continuous authentication failure is determined to be an abnormal behavior caused by limited battery performance.

8. An abnormal behavior recognition system based on a smart door lock, characterized in that, Also includes: Strategy generation module; The strategy generation module is configured to: If the continuous authentication failure is an abnormal behavior caused by illegal intrusion, the alarm module that is connected to the policy generation module is activated to perform alarm processing. When the continuous authentication failure is an abnormal behavior caused by limited battery performance, the TYPE-C interface of the smart lock is activated, and the voice module connected to the policy generation module is started to instruct the initiator of the abnormal behavior to connect the initiator's mobile phone to the TYPE-C interface for data collection and processing required for the authentication of the smart lock based on the initiator's mobile phone.

9. A method for identifying abnormal behavior based on a smart door lock, applied to an abnormal behavior identification system based on a smart door lock as described in any one of claims 1-8, characterized in that, include: During the continuous authentication failure of the smart door lock, the battery terminal voltage sequence and the loop current sequence are collected simultaneously. Identify the initial charge pulses of discrete events in the loop current sequence, construct the initial charge pulse sequence, and obtain the decoupling recovery limitation index based on the differential change trend of the initial charge pulse sequence. Then, derive the equivalent recovery time constant characterizing the recovery capability of the power distribution network from the decoupling recovery limitation index. Using the equivalent recovery time constant, a time-domain filtering mask and energy recovery deduction term are constructed to perform dynamic interference compensation on the battery terminal voltage sequence and voltage-current product data, so as to calculate the voltage recovery hysteresis rate and cumulative energy consumption slope after removing the dynamic interference introduced by the limited recovery of the power distribution network. Using the voltage recovery hysteresis rate and the cumulative energy consumption slope as input parameters, the battery health score is calculated through a particle swarm optimization algorithm model. The battery health score is then compared with a preset judgment threshold to generate an abnormal behavior identification result.

Images