An event recognition method and system based on temporal freezing and heterogeneous decoupling

By using time-series freezing and heterogeneous decoupling, the low-power processing unit collects and stores vibration signals in sleep mode. After waking up, it splices the data with the high-power processing unit, which solves the contradiction between low power consumption and high computing power in long-cycle monitoring terminals and realizes reliable monitoring of long-life batteries and high-power alarm output.

CN123068073APending Publication Date: 2026-11-10SHENZHEN QIAOFU INTELLIGENT EQUIPMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611128864.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-28
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Modern long-cycle monitoring terminals face a contradiction between long standby time and high computing power. Traditional solutions struggle to process complex waveform recognition algorithms under low power consumption, and waking up the high-performance processor causes the battery to be quickly depleted, leading to the loss and misjudgment of key waveforms.

Method used

By employing a time-series freezing and heterogeneous decoupling method, vibration signals are collected and stored by a low-power processing unit in sleep mode. After being woken up, the data is spliced ​​with that of a high-power processing unit to achieve continuous waveform analysis. The high-power processing unit is powered on after the target event is identified.

Benefits of technology

It achieves complete waveform analysis and high-power alarm output under microampere-level standby power consumption, and the long-lasting battery supports reliable monitoring for 10 to 20 years, eliminating data gaps and minimizing energy consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123068073A_ABST
    Figure CN123068073A_ABST
Patent Text Reader

Abstract

This invention provides an event recognition method based on time-series freezing and heterogeneous decoupling, comprising: S1, a detection device in sleep mode collects vibration signals and stores them in a first storage unit in a time sequence; if the detection device recognizes that the vibration signal meets the preset characteristics of an event, it locks the overlay pointer of the first storage unit, controls the newly collected vibration signal to be stored in a second storage unit, and sends a wake-up signal to a low-power processing unit; S2, the low-power processing unit is activated, and a continuous vibration waveform is obtained by splicing a first historical signal, a second transition signal, and a real-time vibration signal; during sleep and signal splicing, the high-power processing unit is forced to maintain a physically powered-off state; S3, the low-power processing unit performs target event recognition based on the continuous vibration waveform, and maintains the physically powered-off state of the high-power processing unit during the target event recognition period; S4, if the low-power processing unit recognizes a target event, it controls a second power supply unit to supply power to the load unit.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of low-power monitoring technology, specifically relating to an event recognition method and system based on time-series freezing and heterogeneous decoupling. Background Technology

[0002] Modern long-cycle monitoring terminals (such as disaster prevention equipment and field sensors) often face a contradiction between "long standby time" and "high computing power". Traditional solutions use a single processor architecture. If low-power chips (such as MCUs) are used, it is difficult to handle complex waveform recognition algorithms. If a high-performance application processor (such as an A55 core running Linux) is activated, its startup current is usually in the tens of milliamps. For long-life batteries (such as Li-SOCI2) with extremely high internal resistance and microamp-level current output, this will deplete the power in a very short time or cause the system voltage to collapse.

[0003] Furthermore, after a sensor triggers an interrupt, there is a physical delay of tens of milliseconds between the main control chip's deep sleep state and its full activation. Existing technologies often directly stop sampling or overwrite old data, resulting in the loss of these tens of milliseconds of critical waveforms (such as the transition segment from P-wave to S-wave in an earthquake), leading to misjudgments. Summary of the Invention

[0004] To address the problems mentioned in the background, this invention discloses an event recognition method and system based on time-series freezing and heterogeneous decoupling. By eliminating data gaps during the wake-up period through "time-series freezing" and combining "heterogeneous computing power and energy decoupling," it achieves compatibility between complete waveform analysis and high-power alarm output under microampere-level standby power consumption, enabling long-life batteries to support reliable monitoring for up to 10 to 20 years.

[0005] A first aspect of the present invention provides an event recognition method based on timing freeze and heterogeneous decoupling, applied to a heterogeneous computing terminal comprising a low-power processing unit and a high-power processing unit, wherein the terminal is heterogeneously powered by a first power supply unit and a second power supply unit, the method comprising: S1, the detection device collects vibration signals in sleep mode and stores them into the first storage unit in sequence. If the detection device recognizes that the vibration signal meets the preset event characteristics, it locks the overlay pointer of the first storage unit, controls the newly collected vibration signal to be stored into the second storage unit, and sends a wake-up signal to the low-power processing unit. S2, the low-power processing unit switches from sleep state to active state in response to the wake-up signal, acquires the first historical signal in the first storage unit, the second transition signal in the second storage unit, and the currently acquired real-time vibration signal, splices the first historical signal, the second transition signal, and the real-time vibration signal to obtain a continuous vibration waveform, and forces the high-power processing unit to maintain a physical power-off state during sleep, wake-up, and signal splicing. S3, the low-power processing unit performs target event identification based on the continuous vibration waveform, and maintains the physical power-off state of the high-power processing unit during the target event identification period; S4. If the low-power processing unit detects the target event, it controls the second power supply unit to supply power to the load unit.

[0006] Furthermore, the first storage unit is the main sequential memory, and the second storage unit is a spare buffer; or, the first storage unit and the second storage unit are different address ranges of the same circular buffer.

[0007] Furthermore, step S1 includes: Step S11: The detection device continuously collects vibration signals in a low-power state and cyclically writes the vibration signals into the first storage unit; Step S12: If the detection device identifies that the vibration signal matches the candidate event characteristics, a wake-up signal is generated; In step S13, in response to the candidate event characteristics, the detection device locks the overwrite pointer of the first storage unit to prevent historical data from being overwritten, and redirects the newly acquired vibration signal to be written to the second storage unit.

[0008] Furthermore, step S13 includes: Step S131: The detection device locks the current safe read boundary pointer in a single circular buffer; In step S132, the detection device, under memory access control, overwrites the oldest data area in the annular buffer with the latest vibration signal.

[0009] Furthermore, step S2 includes: Step S21: During the delay period when the low-power processing unit switches from sleep state to active state, the detection device continuously writes the real-time vibration signal into the second storage unit. In step S22, the low-power processing unit sequentially reads the first historical signal in the first storage unit and the second transition signal in the second storage unit, and receives the real-time vibration signal currently output by the detection device. In step S23, the low-power processing unit seamlessly splices the first historical signal, the second transition signal, and the real-time vibration signal in chronological order to construct a continuous vibration waveform with continuous time.

[0010] Furthermore, step S3 includes: Extract the arrival time difference between the first vibration component and the second vibration component in the continuous vibration waveform. If the time difference conforms to the preset physical propagation law, it is determined to be the target event.

[0011] Furthermore, step S4 includes: during the period when the second power supply unit supplies power to the load unit, the first power supply unit only maintains the microampere-level logic operation of the low-power processing unit and is physically isolated from the power supply circuit of the load unit.

[0012] Furthermore, after step S4, the method further includes: after completing the target event alarm, controlling the detection device to switch to a low threshold monitoring mode; if the amplitude of the external vibration signal reaches the preset monitoring threshold, triggering a response signal using the residual power of the second power supply unit.

[0013] Furthermore, the response signal is used to trigger the second power supply unit to output a short drive current to the load unit.

[0014] A second aspect of the present invention provides an event recognition system based on temporal freezing and heterogeneous decoupling, which uses the method described above to identify events based on temporal freezing and heterogeneous computing decoupling. The system includes: The testing equipment is equipped with a dual-track storage architecture; Heterogeneous computing terminal, comprising a microcontroller as a low-power processing unit and an application processor as a high-power processing unit; The power management module is used to control the first power supply unit to route charging to the second power supply unit in a specific direction; The power supply rail of the application processor is controlled by the microcontroller and remains physically disconnected during sleep, wake-up, signal splicing, target event recognition, and target event alarm.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. Eliminate data gaps: By using "main memory freeze + spare area relay", the sampling blind spot during MCU wake-up is perfectly filled, and the complete physical evolution waveform is obtained.

[0016] 2. Extreme energy consumption compression: The high-power large core (Application Processor) is prohibited from being woken up throughout the entire life cycle. Complex calculations are completed only by the low-power small core (MCU), which enables the long-lasting battery with microampere power supply to support a standby life of 10 to 20 years.

[0017] 3. High-efficiency energy utilization: The second power supply unit (such as a supercapacitor) recovers the depassivation pulse energy of the first power supply unit (such as a chemical battery) and converts it into a high-power alarm output, solving the physical problem that "weak batteries cannot drive strong loads". Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an event recognition method based on temporal freezing and heterogeneous decoupling proposed in this invention; Figure 2 This is a schematic diagram of the structure of an event recognition system based on temporal freezing and heterogeneous decoupling proposed in this invention. Detailed Implementation

[0020] 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.

[0021] The first aspect of the present invention provides an event recognition method based on time-series freezing and heterogeneous decoupling, which is applied to a heterogeneous computing terminal containing a low-power processing unit and a high-power processing unit. The terminal can be heterogeneously powered by a first power supply unit and a second power supply unit. The method achieves continuous monitoring and accurate recognition of vibration events while ensuring low-power operation through a hierarchical storage management mechanism and a heterogeneous processing architecture.

[0022] like Figure 1 As shown, the method includes the following steps: S1, the detection device collects vibration signals in sleep mode and stores them into the first storage unit in sequence. If the detection device recognizes that the vibration signal meets the preset event characteristics, it locks the overlay pointer of the first storage unit, controls the newly collected vibration signal to be stored into the second storage unit, and sends a wake-up signal to the low-power processing unit.

[0023] Optionally, this step also includes: Step S11: The detection device continuously collects vibration signals in a low-power state and cyclically writes the vibration signals into the first storage unit.

[0024] Optionally, the low-power state of the detection device refers to the core processing unit and most functional modules of the device being in a dormant or off state, with only the vibration sensor, analog-to-digital converter, and basic storage control logic remaining active.

[0025] In practical applications, detection equipment is typically deployed in environments requiring long-term monitoring, such as earthquake vibration monitoring points or critical components of disaster recovery terminals. When no abnormal vibrations are detected, the equipment is in a dormant state, during which its main power consumption comes from continuous sampling by the vibration sensor and basic signal buffering operations. The detection equipment integrates a vibration sensor that continuously collects vibration signals from the environment at a fixed sampling frequency, typically set between 1 kHz and 5 kHz, to ensure the complete frequency components of the target vibration event are captured. The collected vibration signals are digitally written sequentially to the first storage unit in chronological order.

[0026] Optionally, the digitized vibration signals are generated at fixed time intervals, with each sampling point containing a timestamp and a corresponding vibration amplitude. These sampling points are sequentially written into the circular buffer of the first storage unit according to the acquisition order. The circular writing mechanism is implemented by maintaining a write pointer, which initially points to the starting address of the buffer. After each sampling point is written, the pointer moves one position forward. When the pointer reaches the end of the buffer, it automatically wraps back to the starting position to continue writing, thus forming a circular data flow path. In one embodiment, the first storage unit adopts a circular buffer storage architecture with a fixed storage capacity. When new vibration data is written, if the buffer is full, the latest data will overwrite the oldest historical data written in the buffer. This circular overwrite mechanism ensures that the first storage unit always stores the vibration signals within the most recent time window. The length of the time window is determined by the buffer capacity and the sampling frequency. For example, if the capacity of the first storage unit is 8,000 sampling points and the sampling frequency is 2 kHz, the buffer can store the vibration history data of the most recent four seconds.

[0027] Step S12: If the detection device identifies that the vibration signal matches the candidate event characteristics, a wake-up signal is generated.

[0028] In one embodiment, while the detection device continuously acquires vibration signals, its internal hardware judgment unit performs rapid feature determination on the real-time acquired signals. The hardware judgment unit is a low-power dedicated circuit module designed to perform preliminary screening of vibration signals with microwatt-level power consumption. The hardware judgment unit makes a matching judgment by comparing the amplitude, frequency, or energy characteristics of the vibration signal with preset event characteristics.

[0029] Optionally, event preset features are typically set for specific types of candidate events.

[0030] For example, let's illustrate the specific form of candidate event characteristics using an earthquake monitoring scenario. When an earthquake occurs, the first wave to arrive is the high-frequency, low-amplitude longitudinal wave (P-wave), whose frequency components are typically concentrated in the range of 1Hz to 20Hz (corresponding to long-period ground motions in geophysics). Background vibrations in the outdoor environment or building structures mainly originate from wind sloshing, human activities, or thermal noise, with frequencies typically dispersed in the low-frequency range of 0.1Hz to 1Hz, or manifesting as high-frequency random noise. Based on this difference in physical characteristics, the candidate event characteristics for an earthquake event can be set as follows: the energy proportion of the vibration signal in the 1Hz to 20Hz frequency band exceeds 50% (distinguishing it from the flat spectrum of background noise), and the peak acceleration in this frequency band exceeds three times the average background vibration (e.g., exceeding 0.05g), while the signal duration is greater than 0.5 seconds to exclude transient impact interference.

[0031] Optionally, the generation of a wake-up signal signifies that the detection device has transitioned from a normal listening state to an event response state. This signal is transmitted to the low-power processing unit via an internal bus or a dedicated interrupt line, triggering the low-power processing unit to switch from sleep mode to active mode.

[0032] In step S13, in response to the candidate event characteristics, the detection device locks the overwrite pointer of the first storage unit to prevent historical data from being overwritten, and redirects the newly acquired vibration signal to be written to the second storage unit.

[0033] In one embodiment, after the hardware determination unit generates a wake-up signal, the storage control logic of the detection device immediately executes a data protection process, specifically locking the overwrite pointer of the first storage unit. The overwrite pointer is an internal index variable in the circular buffer used to indicate the next write position. Locking this pointer means stopping further write operations to the first storage unit, ensuring that the historical vibration data already stored in the first storage unit is completely preserved.

[0034] Meanwhile, the detection equipment needs to continue acquiring new vibration signals to capture the subsequent evolution of the event. To avoid data loss, the detection equipment redirects the newly acquired vibration signals and stores them in a second storage unit.

[0035] The second storage unit is a separate storage area from the first storage unit. Its function is to temporarily store newly generated vibration data during the transition period before the low-power processing unit is woken up.

[0036] Optionally, while completing the storage switch, the detection device notifies the low-power processing unit to prepare to start the high-precision event recognition process.

[0037] In one embodiment, the first storage unit is a primary timing memory, and the second storage unit is a backup buffer.

[0038] In another embodiment, the first storage unit and the second storage unit are different address ranges of the same circular buffer.

[0039] Optionally, this step also includes: In step S131, the detection device locks the current safe read boundary pointer in a single circular buffer.

[0040] In one embodiment, the first and second storage units can share the same physical memory, achieving functional separation through logical partitioning and pointer management. This memory is organized as a single circular buffer with fixed start and end addresses, managed cyclically through the coordinated use of read and write pointers. During normal cyclic writing, the write pointer continuously advances, wrapping back to the start position when it reaches the end of the buffer, thus overwriting the earliest written data.

[0041] When a memory protection operation is triggered, the detection device reads the current write pointer position and saves the value of that position to a dedicated register. The pointer value stored in this register is defined as the safe read boundary pointer. The safe read boundary pointer identifies the boundary between the end of a historical data segment and the beginning of a new data segment. By locking this pointer, the system determines the range that the low-power processing unit should read when subsequently reading historical data, i.e., backtracking from the position pointed to by the safe read boundary pointer to the earliest historical data position that the buffer can provide.

[0042] In step S132, the detection device, under memory access control, overwrites the oldest data area in the annular buffer with the latest vibration signal.

[0043] In one embodiment, after locking the safe read boundary pointer, the write pointer continues to advance to receive newly acquired vibration signals. Since the safe read boundary pointer has marked the boundary of the historical data that needs to be protected, the write pointer can cover the old data areas before the boundary during its advancement. These areas are beyond the historical window range required for the current event analysis and can therefore be safely overwritten.

[0044] Optionally, the total depth of the circular buffer is greater than the sum of the duration of the first historical signal and the wake-up delay of the low-power processing unit.

[0045] Optionally, the total depth of the circular buffer refers to the maximum number of sampling points that the buffer can accommodate.

[0046] Specifically, the total depth of the buffer must be greater than the number of sampling points corresponding to the sum of the duration of the first historical signal and the wake-up delay of the low-power processing unit. The duration of the first historical signal refers to the length of the historical data window required to fully capture the characteristics of the target event, and this duration is determined according to the specific application scenario. For example, in earthquake monitoring, it is necessary to acquire at least 2 seconds of background vibration data before the arrival of the P-wave (first arrival wave) and 6 seconds of data from the arrival of the P-wave to the triggering of the S-wave (infrasound) alarm, in order to fully extract the initiation characteristics and energy accumulation process of the seismic wave. In this case, the duration of the first historical signal is set to 8 seconds.

[0047] The wake-up delay of a low-power processing unit refers to the time interval between the issuance of the wake-up signal and the start of the low-power processing unit to perform data reading operations. This delay includes the power-on stabilization time of the hardware circuit, the startup time of the clock system, and the execution time of the startup program.

[0048] For example, assuming a wake-up delay of 100ms and a sampling frequency of 2kHz, two hundred new sampling points will be generated during the delay period. Therefore, the total depth of the circular buffer needs to accommodate at least 8000 historical sampling points corresponding to four seconds plus 200 transition sampling points corresponding to 100ms, for a total of 8200 sampling points.

[0049] S2, the low-power processing unit switches from sleep state to active state in response to the wake-up signal, acquires the first historical signal in the first storage unit, the second transition signal in the second storage unit, and the currently acquired real-time vibration signal, splices the first historical signal, the second transition signal, and the real-time vibration signal to obtain a continuous vibration waveform, and forces the high-power processing unit to maintain a physical power-off state during sleep, wake-up, and signal splicing.

[0050] Specifically, the low-power processing unit (LDP) is the main control unit responsible for high-precision event recognition in the system. Its hardware typically employs a microprocessor or digital signal processor with strong computing power. During the system's normal monitoring phase, the LDP is in a sleep state to reduce power consumption. Its internal clock system operates at an extremely low frequency or is completely shut down, and peripheral interfaces and processing units are either power-off or gated. When a wake-up signal from the detection device reaches the wake-up pin of the LDP, it triggers the power management module of the LDP to initiate a power-on sequence.

[0051] In one embodiment, the power-on sequence first restores the clock system to normal operation, then sequentially activates the power domains of each functional module, gradually bringing the processor core, memory interface, and arithmetic units into an operational state. After the hardware is ready, the low-power processing unit begins executing startup code. This startup code completes necessary register initialization and peripheral configuration, and finally jumps to the entry address of the event recognition program to begin execution. The entire wake-up process, from signal triggering to program execution, typically takes tens of milliseconds.

[0052] Optionally, this step also includes: In step S21, during the delay period when the low-power processing unit switches from sleep state to active state, the detection device continuously writes the real-time vibration signal into the second storage unit.

[0053] To avoid data loss, the detection device immediately switches the storage target to the second storage unit after the wake-up signal is issued. All newly acquired vibration signals are written to the contiguous address space of the second storage unit in chronological order. The write operation is completed autonomously by the storage control logic of the detection device, without the participation of the low-power processing unit. During this period, the second storage unit acts as a data buffer, temporarily storing all vibration sampling points from the time the wake-up signal is received from the low-power processing unit to the actual completion of the startup process.

[0054] Optionally, the high-power processing unit is physically powered off during sleep and wake-up periods.

[0055] In one embodiment, to minimize system power consumption, the high-power processing unit remains physically powered down during the event recognition cycle. Physical power-down means that all power domains of the high-power processing unit are disconnected, and the processor core, memory, peripheral modules, etc., are all in a zero-power silent sleep state. Compared to software-controlled sleep mode, this physical power-down method can more thoroughly eliminate leakage current, reducing the power consumption of the high-power processing unit to a negligible level.

[0056] When the event recognition process is started, the low-power processing unit confirms that the power control signal of the high-power processing unit is in a physically powered-off state. Only when the target event is confirmed and subsequent response actions need to be initiated will the low-power processing unit activate the power of the high-power processing unit, enabling it to enter the working state and perform the corresponding control or communication tasks.

[0057] This heterogeneous decoupling design strategy allows the system to select the most suitable combination of processing units at different operating stages. During the low-power monitoring stage, it relies solely on low-power detection hardware; in the initial event identification stage, low-power processing units are activated for accurate judgment; and high-power processing units are activated only when complex response actions are required. By activating different processing resources in stages, the system achieves fine-grained power management while ensuring functional integrity.

[0058] In step S22, the low-power processing unit sequentially reads the first historical signal in the first storage unit and the second transition signal in the second storage unit, and receives the real-time vibration signal currently output by the detection device.

[0059] After the low-power processing unit wakes up and enters the active state, it first acquires complete vibration data to support subsequent event recognition and analysis. The complete vibration data consists of signals from three time periods: a first historical signal, a second transition signal, and a real-time vibration signal. The first historical signal originates from the first storage unit and covers vibration records from before the event is initially identified until the hardware judgment unit triggers a response. The second transition signal originates from the second storage unit and records vibration data during the wake-up delay from the issuance of the wake-up signal to the completion of the low-power processing unit's startup. The real-time vibration signal is the current vibration data acquired in real-time from the detection device after the low-power processing unit starts up. The real-time vibration signal is a continuously generated data stream; the low-power processing unit determines the reception duration based on the needs of the event recognition algorithm, typically receiving several seconds of real-time data to ensure sufficient information for accurate judgment.

[0060] The low-power processing unit sequentially reads the cached vibration data from the first and second storage units via an internal bus or dedicated data interface, while simultaneously establishing a data path with the detection device to receive real-time vibration signals. The reading process is performed chronologically to ensure the acquired data is continuous and seamless along the timeline. After acquiring the three signal segments, the low-power processing unit splices these three segments in chronological order within the internal storage space, forming a continuous vibration waveform that extends from before the event to the current moment. This continuous vibration waveform provides a complete data foundation for subsequent feature extraction and event discrimination.

[0061] In step S23, the low-power processing unit seamlessly splices the first historical signal, the second transition signal, and the real-time vibration signal in chronological order to construct a continuous vibration waveform with continuous time.

[0062] Optionally, the high-power processing unit also remains physically powered off during signal splicing.

[0063] In one embodiment, after acquiring three vibration signals, the low-power processing unit needs to combine them into a unified continuous vibration waveform for subsequent processing. The core of the splicing operation is to ensure strict continuity of data on the time axis, avoiding time jumps or data overlap. The low-power processing unit first allocates a continuous buffer in its internal storage space, large enough to accommodate the total length of the three signals. The splicing process fills the buffer sequentially according to the chronological order. First, all sampling points of the first historical signal are copied to the beginning of the buffer according to their original chronological order, with the timestamp corresponding to the last sampling point of the first historical signal recorded as T1. Then, the sampling points of the second transition signal are filled immediately after the first historical signal. The timestamp of the first sampling point of the second transition signal should be equal to T1 plus one sampling period to ensure seamless connection on the time axis. The timestamp of the last sampling point of the second transition signal is recorded as T2.

[0064] Finally, the sampling points of the real-time vibration signal are continued to fill the buffer. The timestamp of the first sampling point of the real-time vibration signal must also be equal to T2 plus one sampling period. Through strict timestamp alignment, the three signal segments form a time-continuous and gapless vibration waveform sequence in the buffer.

[0065] The constructed continuous vibration waveform covers the complete evolution process on the time axis from the background stage before the target event occurs, the event triggering stage, the system wake-up transition stage, to the current real-time stage.

[0066] S3, the low-power processing unit performs target event identification based on the continuous vibration waveform, and maintains the physical power-off state of the high-power processing unit during the target event identification period.

[0067] Optionally, the arrival time difference between the first vibration component and the second vibration component in the continuous vibration waveform can be extracted. If the time difference conforms to the preset physical propagation law, it is determined to be the target event.

[0068] Specifically, target seismic events can be identified based on the temporal evolution of continuous vibration waveforms, their arrival sequence, changes in energy envelope, or propagation characteristics.

[0069] Optionally, the high-power processing unit remains physically powered off during target event identification.

[0070] In one embodiment, target seismic events are identified based on the temporal evolution of continuous vibration waveforms, their arrival order, changes in energy envelope, or propagation characteristics.

[0071] In one embodiment, after obtaining a complete continuous vibration waveform, the low-power processing unit begins the target event identification process. The core task of event identification is to extract characteristic parameters that can characterize the seismic event from the continuous vibration waveform, and to match these characteristic parameters with preset judgment criteria to ultimately determine whether the current vibration signal corresponds to the target event that needs to be responded to.

[0072] Specifically, the continuous vibration waveform in the time domain is a time-varying amplitude sequence, which is actually the combined result of the superposition of seismic wave groups and background noise.

[0073] For example, in an earthquake monitoring scenario, the vibration waveform includes various seismic wave components such as P-waves (Primary Waves), S-waves (Secondary Waves), and surface waves, as well as environmental noise. To accurately identify earthquake events, it is necessary to separate the wave components of different propagation modes. Low-power processing units employ frequency domain filtering or time-frequency analysis techniques to achieve waveform separation.

[0074] Specifically, P-waves propagate at approximately 5–6 km / s in rock strata or building structures, with their frequency components mainly concentrated in the 1 Hz to 20 Hz range (long-period ground motion), exhibiting characteristics of high-frequency micro-tremors and early arrival. S-waves propagate at approximately 3–4 km / s, with their frequency components concentrated in the 0.1 Hz to 10 Hz range, exhibiting characteristics of large amplitude vibrations, strong destructive power, and lagging arrival after the P-wave. The low-power processing unit first performs a Fast Fourier Transform on the continuous vibration waveform to obtain its spectral distribution. Then, two bandpass filters are designed: the first filter has a passband of 1 Hz to 20 Hz for extracting the P-wave component; the second filter has a passband of 0.1 Hz to 10 Hz for extracting the S-wave component. The continuous vibration waveform is passed through these two filters to obtain two filtered output signals. The first output signal is the first vibration component, mainly containing the energy of the P-wave; the second output signal is the second vibration component, mainly containing the energy of the S-wave. Because P-waves and S-waves propagate at different speeds, there is a significant time gap (SP Time Gap) between their arrival times at the monitoring point. This time difference is reflected in the filtered vibration component waveform as the different times when their respective energy peaks occur. Optionally, after extracting the first and second vibration components, it is necessary to determine their arrival times at the monitoring point. The arrival time is usually defined as the moment when the energy first exceeds the detection threshold, or the moment when the energy envelope reaches its peak value. The low-power processing unit first extracts the envelope of the first vibration component. Envelope extraction can be achieved through Hilbert transform or simple sliding window energy calculation. Taking sliding window energy calculation as an example, the low-power processing unit sets a time window, the length of which is usually several sampling periods, such as five sampling points. This window slides along the time axis of the first vibration component, moving one sampling point at a time, and calculating the sum of the squares of the amplitudes of all sampling points within the window as the instantaneous energy at that window position. By traversing the entire first vibration component, an energy envelope curve is obtained. The energy envelope curve smooths the rapid oscillations of the original waveform and highlights the overall trend of energy change.

[0075] The energy peak point is located on the energy envelope curve, and the time corresponding to the peak point is the first arrival time. Peak detection can be achieved by finding local maxima of the curve, while setting certain threshold conditions to exclude weak background fluctuations. For example, the energy at the peak point should be at least five times the mean of the envelope curve, and there should be no higher energy values ​​within a certain range before and after the peak point. With these constraints, the low-power processing unit can accurately locate the arrival time of the most significant energy in the first vibration component, which corresponds to the time when the P-wave packet arrives at the monitoring point.

[0076] Using the same method, the low-power processing unit performs envelope extraction and peak detection on the second vibration component to obtain the second arrival time. The second arrival time corresponds to the time when the torsional wave packet arrives at the monitoring point. Since the propagation speed of the torsional wave is slower than that of the longitudinal wave, the second arrival time is usually later than the first arrival time, and there is a positive time difference between the two.

[0077] After obtaining the first and second arrival times, the low-power processing unit performs a simple subtraction operation to calculate the time difference between them. The formula for calculating the time difference is: the time difference equals the second arrival time minus the first arrival time. Since the torsional wave corresponding to the second arrival time propagates more slowly, its arrival time is later than that of the longitudinal wave, therefore the time difference is a positive value.

[0078] Optionally, the low-power processing unit compares the time difference with a preset time difference threshold range.

[0079] Specifically, the low-power processing unit reads the upper and lower limits of the preset time difference threshold range, and then compares the calculated measured time difference with this range.

[0080] In one embodiment, the comparison operation is achieved through two numerical comparisons. First, it is determined whether the measured time difference is greater than or equal to the lower limit of the time difference threshold range, and then it is determined whether the measured time difference is less than or equal to the upper limit of the time difference threshold range.

[0081] If the measured time difference meets both of the above conditions, it is determined that the time difference is within the time difference threshold range, indicating that the propagation characteristics of the current vibration signal are consistent with the physical evolution of the leakage event. If the measured time difference is less than the lower limit or greater than the upper limit, it is determined that the time difference exceeds the preset range, indicating that the current vibration signal does not conform to the characteristics of the leakage event and may originate from other types of interference or background vibration.

[0082] If the measured time difference meets both of the above conditions, it is determined that the time difference is within the time difference threshold range, indicating that the propagation characteristics of the current vibration signal are consistent with the physical evolution law of seismic waves. If the measured time difference is less than the lower limit or greater than the upper limit, it is determined that the time difference exceeds the preset range, indicating that the current vibration signal does not conform to the characteristics of an earthquake event and may originate from other types of interference or background vibration.

[0083] The comparison process embodies a discrimination logic based on physical laws. Compared to threshold judgment methods that rely solely on vibration amplitude or energy, time difference comparison can more effectively distinguish between seismic events and interference signals. For example, when a terminal is dropped, subjected to violent impact, or during heavy equipment construction, strong vibrations are generated. However, these impact vibrations typically manifest as P-waves and S-waves arriving simultaneously (with a time difference approaching zero), or the waveform does not exhibit the typical evolutionary characteristics of natural earthquakes: "high-frequency, low-amplitude P-waves precede, and low-frequency, high-amplitude S-waves lag behind." Through time difference comparison, the system can eliminate these interference events that, although characterized by high vibration intensity, do not conform to the physical mechanism, significantly reducing the false alarm rate. When the time difference comparison result meets the range conditions, the low-power processing unit confirms that the current continuous vibration waveform corresponds to a real seismic event, completing the identification and judgment of the target event. The identification result is recorded in an internal register in the form of a judgment flag, and simultaneously, the low-power processing unit generates a trigger command. The trigger command is a control signal or data packet, including information such as the event type identifier, identification confidence level, and event occurrence time, used to initiate the subsequent event response process.

[0084] S4. If the low-power processing unit recognizes the target event, it controls the second power supply unit to supply power to the load unit.

[0085] Optionally, while the second power supply unit is supplying power to the load unit, the first power supply unit only maintains the microampere-level logic operation of the low-power processing unit and is physically isolated from the power supply circuit of the load unit.

[0086] In one embodiment, the system employs a dual-power architecture to optimize power management and energy scheduling. The first power supply unit typically uses a high-energy-density primary battery or a long-life energy storage device, designed to provide a continuous and stable power supply to the system's low-power monitoring module over a long operating period. The second power supply unit uses a rechargeable energy storage device, such as a supercapacitor or a secondary battery, to temporarily store electrical energy and provide instantaneous high-current output to high-power loads when needed. This dual-power hierarchical power supply architecture decouples continuous low-power supply requirements from intermittent high-power supply requirements, avoiding the impact damage and efficiency loss to the battery caused by high-power loads directly drawing power from the first power supply unit.

[0087] During normal system operation, the first power supply unit is responsible for powering the detection equipment and low-power processing units. These modules have low and relatively stable power consumption levels, allowing the first power supply unit to operate near its rated operating point for extended periods, maintaining high energy conversion efficiency. Simultaneously, the first power supply unit also needs to periodically supply power to the second power supply unit to maintain its energy storage state. Since the second power supply unit is in standby mode most of the time and does not supply power externally, its energy consumption mainly comes from the self-discharge losses of its components; therefore, the frequency of recharging is relatively low.

[0088] In one embodiment, the low-power processing unit remains in a sleep state for most of the system operation to conserve power. However, it needs to be periodically woken up to perform charging management tasks in order to maintain the power status of the second power supply unit. This periodic wake-up mechanism is implemented using a low-power timer within the low-power processing unit. This timer continues to run during the processing unit's sleep period, maintaining time counting with extremely low power consumption. The timer's counting period is set according to the self-discharge characteristics and charging requirements of the second power supply unit, typically ranging from tens of seconds to several minutes.

[0089] In one embodiment, a timer generates periodic wake-up events at preset time intervals. These wake-up events trigger a low-power processing unit to briefly activate from a sleep state and execute a charging management program. The charging management program first detects the current voltage or power status of the second power supply unit. If the detected power level is lower than a set maintenance threshold, the charging process is initiated. The charging process is implemented by controlling a dedicated power switching device connected in the charging path between the first and second power supply units.

[0090] Optionally, the charging process employs a pulse charging mode instead of a continuous charging mode. Pulse charging mode refers to applying only a short-duration charging current pulse during each charging cycle. After the pulse ends, the charging circuit is disconnected, and the second power supply unit enters an open-circuit resting state. This charging method offers several advantages over continuous trickle charging. Pulse charging effectively reduces polarization during the charging process, allowing the concentration gradient within the electrolyte to diffuse and recover during the pulse intervals, reducing overpotential losses at the end of charging and improving charging efficiency. Pulse charging also reduces the instantaneous power demand of the charging current on the first power supply unit, preventing a drop in the output voltage of the first power supply unit during charging that could affect the normal operation of other loads.

[0091] In one embodiment, after the charging circuit is established, the low-power processing unit initiates pulse generation control logic and sends a control signal to the charging switch according to preset pulse parameters. The control signal is a periodic pulse waveform, where the high-level duration of the pulse corresponds to the on-time of the charging current, and the low-level duration of the pulse corresponds to the off-time of the charging current. By adjusting the duty cycle of the pulse width modulation signal, the low-power processing unit can flexibly control the average electrical energy transmitted per unit time.

[0092] Optionally, the low-power processing unit continuously monitors the voltage change of the second power supply unit during the charging process. When the voltage reaches the preset charging termination voltage, it stops sending control pulses, and the charging process ends.

[0093] Through periodic pulse charging management, the system ensures that the second power supply unit is always maintained at a high power level, ready to provide sufficient energy support for responding to sudden events. Compared with passive trickle charging or unmanaged placement, this proactive energy management strategy can more effectively utilize the limited energy of the first power supply unit and extend the overall service life of the system.

[0094] Optionally, after the low-power processing unit completes the identification and determination of the target event and generates a trigger command, the system enters the event response phase. The event response phase requires the load unit to execute a predetermined action. The load unit is typically a high-power actuator or communication module, whose instantaneous power consumption is much higher than the system's daily monitoring power consumption. To avoid damage to the first power supply unit caused by the large current surge during load startup and to ensure power supply stability, the system uses a second power supply unit as the direct power source for the load.

[0095] In one embodiment, after identifying a target event, the low-power processing unit first checks the power status of the second power supply unit to confirm that its stored energy is sufficient to support the operating requirements of the load. If the power is insufficient, the low-power processing unit can choose to perform a quick recharge or adopt a backup emergency response scheme, such as activating only a low-power local alarm without activating the wireless communication module. If the power is sufficient, the low-power processing unit closes the power supply relay or electronic switch connecting the second power supply unit and the load unit via a control signal. After the power supply relay is closed, the output terminal of the second power supply unit is connected to the power supply input terminal of the load unit, and the second power supply unit begins to release drive current to the load unit.

[0096] Powered by the second power supply unit, the system can provide sufficient current output and a stable output voltage. The internal resistance of the supercapacitor is typically in the tens to hundreds of milliohms range; even with an output current of several amperes, the voltage drop across the internal resistance is only a few tenths of a volt, having a minimal impact on the output voltage. The energy stored in the second power supply unit is gradually released during load operation, causing the terminal voltage to decrease. When the terminal voltage drops to the load's minimum operating voltage, the load automatically stops operating, or the low-power processing unit actively disconnects the power supply switch, ending the event response process.

[0097] In one embodiment, before sending the relay control signal, the low-power processing unit first performs a final confirmation of the power status of the second power supply unit. By collecting the terminal voltage of the second power supply unit, it determines whether its stored energy is sufficient to support the expected operating time of the load. If the voltage value reaches the minimum voltage required for load startup and the stored energy meets the product requirement of load power consumption and operating time, then the power supply operation is confirmed to be sufficient, and the power supply operation continues.

[0098] In one embodiment, during the operation of the load unit, the first power supply unit is bypassed and configured to supply power only to the low-power processing unit and the detection device, thereby cutting off the direct connection loop between the first power supply unit and the load unit.

[0099] Specifically, the second power supply unit is responsible for powering the load unit independently during startup and operation, while the first power supply unit is configured to be dedicated to maintaining power for the low-power monitoring circuit during this period.

[0100] When the system enters the event response phase and activates the second power supply unit to supply power to the load, the low-power processing unit disconnects the connection switch from the first power supply unit to the load power supply branch via a control signal. This configuration avoids the impact load on the first power supply unit caused by the high current of the load. During high-current discharge, primary batteries experience significant internal resistance voltage drop and voltage decrease, reducing discharge efficiency and capacity utilization. Continuous high-current discharge also accelerates the battery aging process and shortens battery life. By delegating the load power supply task to the second power supply unit, the first power supply unit only needs to maintain low-power power supply for the low-power processing unit and detection equipment.

[0101] Specifically, during the operation of the load unit, the low-power processing unit and the detection equipment still need to continue working to monitor subsequent changes in vibration signals. The detection equipment continues to collect vibration signals and write them to the storage unit, while the low-power processing unit can continue to analyze the vibration waveform to determine the evolution trend of the event or wait for the load operation to end, depending on the application requirements. The power consumption of these monitoring tasks is relatively stable and low, and the first power supply unit can continuously and stably provide the required power.

[0102] Specifically, after the load unit completes its initial alarm or communication task, the system enters the subsequent monitoring phase. The purpose of this phase is to continuously monitor changes in the vibration state at the site, so as to determine whether the load response needs to be triggered again based on the evolution of the event. To extend the monitoring time with limited remaining power, the low-power processing unit controls the detection device to switch from normal event recognition mode to low-threshold monitoring mode.

[0103] The low-threshold monitoring mode is a lower-power operating mode with adjusted sensitivity. In this mode, the vibration sensor of the detection device remains active and continues to collect vibration signals, but the signal processing and judgment logic is simplified. The detection device no longer performs complex waveform analysis and feature extraction, but only performs a simple threshold comparison of the amplitude or energy of the vibration signal. The monitoring threshold is set below the threshold corresponding to the preset feature of the event that triggers the initial event recognition, enabling the system to maintain responsiveness to weak vibration disturbances.

[0104] The low-power processing unit switches operating modes by sending mode-switching commands to the detection device. These commands can be configuration commands transmitted via a serial communication interface or level signals sent via a dedicated control signal line. Upon receiving the command, the detection device updates its internal operating parameter configuration, lowering the comparison threshold of the hardware judgment unit from the initial trigger threshold to the monitoring threshold. For example, in an earthquake monitoring and post-disaster search and rescue scenario, after the main shock alarm ends, aftershocks may occur at the disaster site, or survivors may call for help by knocking on walls, pipes, etc. The intensity of aftershocks may be lower than the main shock, and the distress signal will appear as short, weak pulse vibrations. By setting a lower monitoring threshold, the system can continuously monitor environmental vibrations with higher sensitivity and promptly capture these subsequent key vibration changes. Once a weak vibration matching the characteristics (such as a knocking sound) is detected, the system uses the residual power from the second power supply unit to trigger a short response signal (such as an intermittent audible and visual alarm), thereby assisting rescue personnel in locating the source of the problem or triggering a load response to update the disaster status.

[0105] The specific value of the monitoring threshold needs to balance sensitivity and anti-interference capability. Setting the threshold too low will cause the system to respond to weak background noise or accidental disturbances, increasing the probability of false triggers and consuming valuable remaining power. Setting the threshold too high may miss important event evolution signals, reducing the system's subsequent monitoring effectiveness. In practical applications, the monitoring threshold is usually set to 50% to 70% of the initial trigger threshold, maintaining a certain level of sensitivity while providing a reasonable margin for anti-interference.

[0106] In low-threshold monitoring mode, the power consumption of the detection device is reduced compared to normal operating mode. This reduction in power consumption is primarily due to the decreased complexity of signal processing; the simplified threshold comparison logic requires fewer computational resources and shorter processing time compared to complete waveform analysis and feature extraction algorithms. The detection device can use a lower operating clock frequency or disable some non-essential functional modules, further reducing power consumption. This power optimization allows the system to maintain subsequent monitoring for a longer period even with limited remaining power in the second power supply unit.

[0107] After the detection equipment switches to low-threshold monitoring mode, the vibration sensor continues to sample environmental vibrations at the set sampling frequency. The sampled vibration signal is input to the hardware judgment unit in digital form, and the hardware judgment unit performs real-time judgment on the signal according to the new monitoring threshold parameters.

[0108] In one implementation, the detection signal directly triggers the second power supply unit to output a short drive current to the load.

[0109] In another implementation, the detection signal first notifies the low-power processing unit for secondary confirmation and determination, and the low-power processing unit decides whether to initiate load response.

[0110] Optionally, after step S4, the method further includes: after completing the target event alarm, controlling the detection device to switch to a low-threshold monitoring mode; if the amplitude of the external vibration signal reaches a preset monitoring threshold, triggering a response signal using the residual power of the second power supply unit. In one embodiment, the response signal is used to trigger the second power supply unit to output a short drive current to the load unit.

[0111] When the detection equipment detects an external vibration signal reaching the monitoring threshold in low-threshold monitoring mode, the system determines that a noteworthy new change has occurred in the field situation and requires the load unit to be restarted to respond. At this time, the second power supply unit has partially consumed its stored energy after continuously supplying power to the load in the previous stage, and the remaining energy may not be sufficient to support the load to work for the same duration as the initial response. In order to make full use of the remaining energy and ensure that the load can complete the necessary response actions, the system adopts a short-drive power supply strategy.

[0112] Upon receiving the monitoring trigger signal from the detection device, the low-power processing unit first assesses the current power status of the second power supply unit. By acquiring the terminal voltage of the second power supply unit and calculating the remaining energy based on the relationship between voltage and stored energy, it determines whether the remaining energy is sufficient to support the load's short-term operation. If the remaining voltage is higher than the load's minimum operating voltage and the remaining energy meets the requirements for short-term operation, the low-power processing unit decides to execute a short-term drive response.

[0113] A short drive current refers to a short-duration drive current pulse that is output to the load. The pulse duration is determined based on the functional characteristics of the load and its remaining electrical energy.

[0114] For example, taking an audible and visual alarm as an example, a short-actuated signal can be an intermittent alarm signal lasting one to three seconds. Compared to a continuous alarm that may last for more than ten seconds during the initial response, a short-actuated signal completes the alert function in a shorter time, significantly reducing energy consumption. For wireless communication modules, a short-actuated signal can be achieved by sending only a single status update data packet without multiple retransmissions or handshake confirmations, reducing energy consumption during communication.

[0115] The low-power processing unit achieves short-duration drive timing control by controlling the on / off state of power supply relays or electronic switches. The control signal is output in pulse form, and the high-level duration of the pulse corresponds to the operating duration of the load. After the pulse ends, the switch opens, and the second power supply unit stops supplying power to the load. The load automatically stops working after completing the predetermined action during the short power supply period, waiting for the next trigger.

[0116] Optionally, the monitoring threshold can be set lower than the trigger threshold corresponding to the preset characteristics of the event. This differentiated threshold design reflects the system's differentiated processing strategy for vibration signals at different stages. A higher initial trigger threshold requires the vibration signal to have obvious target event characteristics before initiating the complete identification and response process, avoiding misjudgment of weak signals. A lower monitoring threshold allows the system to remain sensitive to weaker subsequent vibration changes, promptly capturing the dynamic evolution of the event. The combined use of these two threshold levels ensures the accuracy of initial identification while also considering the comprehensiveness of subsequent monitoring.

[0117] In one specific implementation, the initial trigger threshold is set to a peak vibration acceleration greater than 0.5g (gravitational acceleration), and the listening threshold is set to a peak vibration acceleration greater than 0.3g. Actual measurements show that the background vibration level of an urban environment or building structure under normal conditions (such as pedestrian traffic, distant vehicle traffic, and light breezes) is approximately 0.1g. However, during a destructive earthquake, the peak acceleration of the initial P-wave can reach 0.6g to 1.0g, and the peak values ​​of the subsequent S-wave or surface wave mainshock far exceed this range. The initial trigger threshold can effectively distinguish between earthquake P-waves and everyday background vibrations; while in the low-power listening mode after the mainshock alarm ends, the listening threshold can capture low-intensity aftershock signals, or distress signals from survivors in the 0.3g to 0.5g range, such as knocking or shaking.

[0118] By employing short-duration drive responses, the system maintains its ability to respond promptly to on-site dynamics even with limited remaining power, providing continuous information updates for emergency response and remote monitoring. This on-demand energy allocation strategy fully leverages the energy storage potential of the second power supply unit, maximizing the system's continuous operating time while ensuring critical response functions.

[0119] In another implementation, the system can dynamically adjust the response intensity of short-duration drives based on the amount of remaining power. When there is sufficient remaining power, the duration of the short-duration drive can be appropriately extended, allowing the load to complete a more complete functional output. When the remaining power is nearly depleted, the duration of the short-duration drive is compressed to a minimum, and the load only performs the most essential indicative actions, such as a visual alarm flashing only once or emitting a short warning sound. This adaptive energy management strategy requires a low-power processing unit to make decisions based on real-time power status, determining the appropriate drive duration corresponding to the current power level through table lookup or calculation.

[0120] After completing all short-duration drive responses or when the second power supply unit runs out of power, the system can choose to enter a deep sleep state to wait for external charging or power replacement, or it can continue to rely on the first power supply unit to maintain the most basic vibration monitoring functions, waiting for on-site personnel to arrive and handle the situation. When making state switching decisions, the low-power processing unit comprehensively considers factors such as the current event status, remaining power, and communication status to select the most reasonable subsequent working mode.

[0121] like Figure 2 As shown, in a second aspect, the present invention provides an event recognition system based on time-series freezing and heterogeneous decoupling. The system uses the method described above to identify events based on time-series freezing and heterogeneous computing decoupling. The system includes a detection device, a heterogeneous computing terminal, a power management module, and a power supply unit. Each module achieves high reliability monitoring under extremely low power consumption through hardware-level decoupling design.

[0122] Optionally, the system includes: The testing equipment is equipped with a dual-track storage architecture; Heterogeneous computing terminal, comprising a microcontroller as a low-power processing unit and an application processor as a high-power processing unit; The power management module is used to control the first power supply unit to route charging to the second power supply unit in a specific direction; The power supply rail of the application processor is controlled by the microcontroller and remains physically disconnected during sleep, wake-up, signal splicing, target event recognition, and target event alarm.

[0123] Optionally, the detection device is preferably a MEMS accelerometer sensor, which is internally configured with physical hardware featuring a dual-track storage architecture, specifically including: First storage unit (main FIFO): Used to cyclically store vibration data under normal monitoring conditions. When the hardware state machine inside the sensor detects that the vibration signal exceeds the preset initial trigger threshold (e.g., 0.5g), it immediately locks the cover pointer of the first storage unit through the hardware latch pin.

[0124] The second storage unit (backup safety buffer / shadow register): a physical storage area independent of the main FIFO. Once the lock is triggered, newly acquired real-time vibration signals are seamlessly redirected and written to this area. This design ensures that data is not lost during the tens of milliseconds of "blind zone" when the system's main control chip (MCU) wakes up from deep sleep, completely eliminating waveform discontinuities.

[0125] Optionally, the heterogeneous computing terminal adopts an asymmetric multi-core architecture, containing two types of processing units, and the two are completely independent in terms of power supply: Low-power processing unit (MCU): such as the ARM Cortex-M33 core. This unit remains in standby mode most of the time, connects to detection devices via SPI / I2C bus to read stored data, and connects to a power management module to output control signals. Its power consumption at full speed is only in the milliamp range.

[0126] High-power processing unit (application processor / AP): such as an ARM Cortex-A55 core, running Linux or Android operating systems. In the system of this invention, the power rail of this unit is physically connected in series with an electronic switch composed of MOSFETs. The gate of this switch is directly controlled by the GPIO pin of the low-power processing unit (M33) or the enable terminal of the power management module (PMIC). It is physically power-off for more than 99% of its lifespan. It is only temporarily powered by the MCU when non-real-time big data analysis tasks need to be performed (such as post-event waveform upload to the cloud). Throughout the entire lifespan of earthquake monitoring, the unit remains power-off, thereby eliminating the backlash of hundreds of milliamps of power consumption caused by operating system startup.

[0127] Optionally, the system employs a heterogeneous energy dispatch strategy, with the power management module (PMIC) performing precise routing control: The first power supply unit (long-term energy storage) uses a non-rechargeable Li-SOCI2 battery (such as ER14505). Its characteristics are extremely high energy density and extremely low self-discharge (supporting 10-20 years of standby), but its internal resistance is extremely high and it cannot withstand large current surges.

[0128] The second power supply unit (transient energy storage) uses a supercapacitor bank, characterized by extremely low internal resistance and fast charging and discharging speed.

[0129] Targeted charging mechanism: The PMIC is equipped with a first power switch (such as a Buck-Boost controller), which is controlled by the MCU and periodically turns on the charging circuit from the first power supply unit to the second power supply unit. The pulsed high current during charging tears the passivation film of the battery plates (removing passivation), and at the same time, 100% of the pulse energy is recovered into the supercapacitor, turning waste into treasure.

[0130] Load-driven isolation: When the alarm is triggered, the PMIC is equipped with a second power switch (high-side switch). The PMIC forcibly disconnects the connection between the first power supply unit and the high-power audible and visual alarm (load unit). Instead, the second power supply unit (supercapacitor) is directly connected to the load unit through a low-impedance trace. This high-current power supply path of "supercapacitor directly connected to load" can release a large current (such as 2A) instantly, thereby preventing the weak battery from being "killed".

[0131] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0132] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An event recognition method based on time-series freezing and heterogeneous decoupling, applied to a heterogeneous computing terminal containing low-power processing units and high-power processing units, characterized in that, The method includes: S1, the detection device collects vibration signals in sleep mode and stores them into the first storage unit in sequence. If the detection device recognizes that the vibration signal meets the preset event characteristics, it locks the overlay pointer of the first storage unit, controls the newly collected vibration signal to be stored into the second storage unit, and sends a wake-up signal to the low-power processing unit. S2, the low-power processing unit switches from sleep state to active state in response to the wake-up signal, acquires the first historical signal in the first storage unit, the second transition signal in the second storage unit, and the currently acquired real-time vibration signal, splices the first historical signal, the second transition signal, and the real-time vibration signal to obtain a continuous vibration waveform, and forces the high-power processing unit to maintain a physical power-off state during sleep, wake-up, and signal splicing. S3, the low-power processing unit performs target event identification based on the continuous vibration waveform, and maintains the physical power-off state of the high-power processing unit during the target event identification period; S4. If the low-power processing unit detects the target event, it controls the second power supply unit to supply power to the load unit.

2. The method as described in claim 1, characterized in that, The first storage unit is the main sequential memory, and the second storage unit is the spare buffer; or, the first storage unit and the second storage unit are different address ranges of the same circular buffer.

3. The method as described in claim 2, characterized in that, Step S1 includes: Step S11: The detection device continuously collects vibration signals in a low-power state and cyclically writes the vibration signals into the first storage unit; Step S12: If the detection device identifies that the vibration signal matches the candidate event characteristics, a wake-up signal is generated; In step S13, in response to the candidate event characteristics, the detection device locks the overwrite pointer of the first storage unit to prevent historical data from being overwritten, and redirects the newly acquired vibration signal to be written to the second storage unit.

4. The method as described in claim 3, characterized in that, Step S13 includes: Step S131: The detection device locks the current safe read boundary pointer in a single circular buffer; In step S132, the detection device, under memory access control, overwrites the oldest data area in the annular buffer with the latest vibration signal.

5. The method as described in claim 1, characterized in that, Step S2 includes: Step S21: During the delay period when the low-power processing unit switches from sleep state to active state, the detection device continuously writes the real-time vibration signal into the second storage unit. In step S22, the low-power processing unit sequentially reads the first historical signal in the first storage unit and the second transition signal in the second storage unit, and receives the real-time vibration signal currently output by the detection device. In step S23, the low-power processing unit seamlessly splices the first historical signal, the second transition signal, and the real-time vibration signal in chronological order to construct a continuous vibration waveform with continuous time.

6. The method as described in claim 1, characterized in that, Step S3 includes: Extract the arrival time difference between the first vibration component and the second vibration component in the continuous vibration waveform. If the time difference conforms to the preset physical propagation law, it is determined to be the target event.

7. The method as described in claim 1, characterized in that, Step S4 includes: during the period when the second power supply unit supplies power to the load unit, the first power supply unit only maintains the microampere-level logic operation of the low-power processing unit and is physically isolated from the power supply circuit of the load unit.

8. The method as described in claim 1, characterized in that, After step S4, the method further includes: after completing the target event alarm, controlling the detection device to switch to a low threshold monitoring mode; if the amplitude of the external vibration signal reaches the preset monitoring threshold, using the residual power of the second power supply unit to trigger a response signal.

9. The method as described in claim 8, characterized in that, The response signal is used to trigger the second power supply unit to output a short drive current to the load unit.

10. An event recognition system based on temporal freezing and heterogeneous decoupling, characterized in that, The system employs the method described in any one of claims 1 to 9 to identify events based on temporal freezing and heterogeneous computing decoupling, the system comprising: The testing equipment is equipped with a dual-track storage architecture; Heterogeneous computing terminal, comprising a microcontroller as a low-power processing unit and an application processor as a high-power processing unit; The power management module is used to control the first power supply unit to route charging to the second power supply unit in a specific direction; The power supply rail of the application processor is controlled by the microcontroller and remains physically disconnected during sleep, wake-up, signal splicing, target event recognition, and target event alarm.