Low-power wireless sensor network gateway and control method thereof

Abstract

This invention discloses a low-power wireless sensor network gateway and its control method, relating to the field of gateway control technology. The gateway includes a wireless public network communication unit, a data processing and control unit, a low-power wake-up radio frequency unit, a wireless personal area network radio frequency unit, and a battery and power management module. The data processing and control unit adopts a high- and low-power dual-core architecture. The low-power resident main control unit is always on standby, responsible for battery status monitoring and wake-up command parsing, while the edge computing and power co-processing unit is woken up on demand. The co-processing unit has a built-in joint decision engine that integrates data urgency, battery voltage, battery stability, and device operation prediction information to generate a hierarchical fault-tolerant power management strategy. High-urgency events are uploaded immediately or buffered and re-uploaded under low voltage conditions, while low-urgency data is uploaded in batches. This invention also features predictive optimization and low-power remote wake-up functions, ensuring the real-time nature of critical events while minimizing average power consumption and extending battery life.

Description

Technical Field

[0001] This invention relates to the field of gateway control technology, specifically a low-power wireless sensor network gateway and its control method. Background Technology

[0002] In wireless sensor networks, wireless gateways enable communication between the wireless personal area network (PAN) and the public network on the wireless sensor side, as well as the management of the wireless sensors. Traditional wireless gateways consume a lot of power because they need to connect to both the PAN and the public network in real time, thus typically requiring continuous external power. However, in many industrial monitoring scenarios, environmental limitations or high cabling costs prevent the provision of external power to the gateway, necessitating battery power. Therefore, minimizing the power consumption of wireless gateways and extending their operating time under battery power conditions is a pressing issue in this field.

[0003] The power consumption of a wireless gateway mainly consists of public network communication power consumption and system operation power consumption. Some existing low-power solutions save power by shutting down the public network communication module when not uploading data. For example, Chinese patent CN118041723A discloses a wireless gateway and its control method. However, its decision-making logic is usually simple and fixed rules, determining when to activate the communication module based solely on a time period or a single data matching result. This approach fails to consider the service value of the data in conjunction with the gateway's own energy state. When the gateway's battery is extremely low, reporting non-urgent routine inspection data may cause a system power outage, thus missing potential emergencies. Conversely, in emergency situations such as detecting device malfunctions, it should not be limited by preset upload cycles or power constraints. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies, this invention provides a low-power wireless sensor network gateway and its control method. By employing dual-core collaborative division of labor at the hardware level and multi-dimensional joint judgment at the decision level, it solves the problem of separation between data upload decision and energy state management in existing solutions, thereby maximizing battery life while ensuring real-time reporting capabilities for critical events.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A low-power wireless sensor network gateway includes a wireless public network communication unit, a data processing and control unit, a low-power wake-up radio frequency unit, a wireless personal area network (PAN) radio frequency unit, a battery, and a power management module. The wireless public network communication unit is used to establish public network data communication with a remote server and is in a completely power-off sleep state by default. The PAN radio frequency unit is used to establish local data communication with wireless sensors and is continuously in a low-power listening state to collect sensor monitoring data. The low-power wake-up radio frequency unit has lower static power consumption than the wireless public network communication unit, is always in a low-power listening state, and is used to receive remote wake-up commands from the remote server.

[0007] The data processing and control unit includes a low-power resident master unit and an edge computing and power co-processing unit. The low-power resident master unit operates continuously with power on, monitoring battery power status, battery voltage fluctuations, execution cycle timing, sleep management, and wake-up command parsing in real time. The edge computing and power co-processing unit is a high-performance computing unit that only wakes up to run on demand and is completely powered off after task completion. A dedicated data interaction interface is established between the low-power resident master unit and the edge computing and power co-processing unit. This interface allows the edge computing and power co-processing unit to batch and synchronously transmit power monitoring data cached in the low-power resident master unit during standby, including battery voltage, voltage stability, voltage fluctuations, and static recovery status, after the edge computing and power co-processing unit is woken up and powered stably. The edge computing and power co-processing unit performs edge computing on the collected monitoring data to obtain feature data, determines the urgency of data to be uploaded based on feature data matching results, receives battery power status data transmitted from the low-power resident master unit, and generates a joint power management strategy based on the battery power status. The power management module is electrically connected to the wireless public network communication unit, the data processing control unit, and the battery, respectively, and is used to execute a joint power management strategy to perform fine-grained power-on and power-off control on each functional unit. The battery is used to provide operating power for all gateway hardware units.

[0008] Furthermore, the edge computing and power co-processing unit integrates an edge computing module, a power status awareness module, a joint decision engine module, and a strategy output module. The edge computing module extracts and processes features from raw monitoring data to obtain characteristic data representing the operating status of the device under test. The power status awareness module works with the low-power resident master control unit to receive and parse information on remaining battery power, battery terminal voltage, instantaneous power consumption, and battery voltage fluctuations and stability. The joint decision engine module incorporates a multi-dimensional joint decision model based on data urgency and battery status. It combines four dimensions—data service urgency, remaining battery power, battery physical stability, and device operation prediction information—to generate optimal power consumption control and data upload strategies. Device operation prediction information is used to correct data urgency weights, dynamically fine-tune voltage judgment thresholds, and participate in the entire hierarchical fault-tolerant decision-making process. The strategy output module converts the joint strategy generated by the decision model into hardware control commands and sends them to the power management module for execution. In the multi-dimensional joint decision model, when the remaining power percentage reported by the low-power resident master control unit conflicts with the energy state indicated by the measured terminal voltage, the measured terminal voltage result is used as the priority for decision-making.

[0009] Furthermore, the joint decision engine module is configured with a hierarchical fault-tolerant decision-making mechanism, specifically including high-urgency data processing strategies, low-urgency data processing strategies, and abnormal data dynamic adjustment strategies. For high-urgency event data corresponding to device startup and shutdown, a battery voltage undervoltage protection threshold is set. When the battery voltage exceeds the undervoltage protection threshold, the wireless public network communication unit is immediately triggered to upload data, and power is cut off immediately upon completion of the upload. When the battery voltage falls below the undervoltage protection threshold, high-priority data is encrypted and cached. Data is re-uploaded after the open-circuit voltage caused by the battery polarization voltage decay has been allowed to recover to the preset safe startup voltage threshold and the voltage fluctuation has stabilized. This recovery process is an electrochemical polarization recovery phenomenon and does not represent an increase in the percentage of remaining battery capacity; the re-upload condition is based on the absolute voltage threshold. When device operation prediction information anticipates an impending start-up or shutdown event, the urgency weight of the corresponding event data is proactively increased, and power resources are reserved preferentially. For low-urgency routine monitoring data corresponding to normal device operation, a terminal voltage safety threshold and hysteresis judgment mechanism are configured. Battery false voltage and voltage stability status are continuously monitored and judged by the low-power resident main control unit and pushed in real time. Batch uploads are performed only when the preset upload cycle is reached, the battery voltage is above the safety threshold, and the battery voltage is stable without significant fluctuations or false voltage. At other times, data is temporarily stored and the device remains in low-power sleep mode. If the device is expected to operate in a long-term steady-state state, the periodic upload threshold is appropriately relaxed to further reduce the wake-up frequency. For high-urgency abnormal data corresponding to device malfunctions, after immediate upload, the sensor data acquisition frequency and edge computing execution frequency are dynamically adjusted based on the degree of abnormal deviation. An energy safety constraint is set in conjunction with the battery voltage status; when the voltage is low, the frequency modulation amplitude is reduced to balance monitoring accuracy and device battery life. When predicting the evolution trend of device failures, the priority of abnormal monitoring is increased in advance.

[0010] Furthermore, the edge computing and power co-processing unit also incorporates a time-series learning and predictive optimization module. This module continuously records the time-series historical data of the device under test's start-up, shutdown, and abnormal operation. Through time-series mining, it learns the device's periodic operating patterns, generates device operation prediction information, and inputs it in real-time to the joint decision engine module for multi-dimensional joint decision-making. The time-series learning and predictive optimization module shares a feature sample library stored in the storage unit with the edge computing module. This sample library is updated online on the edge computing module side based on a weighted moving average mechanism to avoid misjudgments caused by aging fixed feature templates. The time-series learning and predictive optimization module is configured with a prediction confidence grading mechanism and a timeout fault-tolerant rollback mechanism. For high-urgency predicted events, low-confidence predictions only preload algorithm parameters, while high-confidence predictions pre-wake up the hardware unit in stages. If no corresponding device event is detected within the prediction window, the preloaded resources are immediately cleared and deep sleep is restored to avoid unnecessary power consumption.

[0011] Furthermore, when the gateway is in deep sleep mode, the low-power resident master control unit maintains microampere-level standby, while the low-power wake-up radio unit and the wireless personal area network radio unit maintain milliampere-level low-power monitoring. After receiving a remote wake-up command, the low-power resident master control unit independently performs command parsing and validity verification. Upon successful verification, it wakes up the edge computing and power coprocessing unit and the wireless public network communication unit as needed to perform remote data interaction or firmware upgrade tasks. After the task is completed, the high-power unit is immediately shut down.

[0012] The second aspect of this invention provides a control method for a low-power wireless sensor network gateway, applicable to any of the aforementioned wireless sensor network gateways, comprising the following steps: S1. The gateway operates in standby mode, with a low-power resident main control unit monitoring the battery status, battery voltage stability and false voltage status in real time, and executing operation cycle timing. The wireless personal area network radio frequency unit continuously collects sensor monitoring data, and the low-power wake-up radio frequency unit continuously listens for remote wake-up commands. S2. When sensor monitoring data is acquired or a preset upload cycle is reached, the low-power resident main control unit prioritizes completing the basic end voltage and voltage status pre-judgment. Only when the basic wake-up conditions are met, the edge computing and power co-processing unit is woken up as needed to perform edge computing feature extraction of monitoring data and accurate verification of battery power status in parallel. S3. The extracted feature data is matched with pre-stored device operation feature sample data to determine the urgency of the current data. Combining the four-dimensional parameters of battery voltage, battery stability, and device operation prediction information, a multi-dimensional joint power management and data upload strategy is generated. S4. Execute corresponding control strategies based on the hierarchical fault-tolerant decision-making mechanism. High-urgency data is uploaded immediately on demand or cached and re-uploaded with priority; low-urgency data is uploaded in batches at different times; abnormal data is dynamically adjusted in terms of acquisition and processing frequency with energy safety net constraints; and prediction results are dynamically adjusted to correct decision weights, achieving deep integration of prediction and hierarchical decision-making. S5. After all data processing, uploading, and device interaction tasks are completed, immediately cut off the power supply to the edge computing and power co-processing unit and the wireless public network communication unit, leaving only the low-power unit in standby mode, and iteratively execute the monitoring and control process.

[0013] Furthermore, the joint strategy generation process in step S3 specifically includes: when the feature data matches the device startup sample or shutdown sample, it is determined to be a high-urgency event, and the terminal voltage limit threshold protection strategy is executed to avoid device downtime and data loss caused by low-voltage high-current impact. Combined with the device's predicted start-up and shutdown trend, the event priority is increased in advance. When the feature data matches the device's normal operation sample, it is determined to be low-urgency routine data. Combined with the terminal voltage hysteresis judgment, battery false voltage stability judgment, and batch upload mechanism, the power consumption oscillation caused by battery false voltage is suppressed to avoid batch high-power transmission depleting battery power. The upload cycle threshold is dynamically optimized based on the steady-state operation prediction result. When the feature data does not match the normal operation sample, it is determined to be a device abnormal event. Abnormal data is uploaded immediately, and the monitoring frequency is dynamically adjusted according to the abnormality level and battery remaining capacity to achieve a dynamic balance between monitoring accuracy and battery life. Combined with the abnormal trend prediction, the monitoring strategy is strengthened in advance.

[0014] Furthermore, it also includes predictive low-power optimization steps: continuously accumulating device runtime sequence data to mine the time patterns of fixed device start-up and shutdown and abnormal occurrences; predicting future device state change nodes based on historical operating patterns, performing pre-start preparation operations based on confidence level classification, and inputting the prediction results into a hierarchical decision-making mechanism in real time to correct decision weights and thresholds; when the prediction timeout fails to trigger the corresponding device event, actively reverting to a deep sleep state to eliminate the invalid power consumption caused by prediction errors; the feature sample library is updated online according to a weighted moving average mechanism to avoid misjudgments caused by the aging of fixed feature templates.

[0015] Furthermore, it also includes a remote low-power wake-up step: In the deep sleep state of the gateway, only the low-power resident master control unit maintains microampere-level standby, while the low-power wake-up radio frequency unit and the wireless personal area network radio frequency unit maintain milliampere-level low-power listening state; after receiving the server wake-up command, the low-power resident master control unit completes the command parsing and verification, wakes up the high-power unit as needed to execute the remote task, and immediately returns to the sleep state after the task is completed, minimizing the overall power consumption of the remote wake-up scenario.

[0016] Compared with existing technologies, this invention has the following significant advantages: First, it clearly defines the dedicated data interaction path for high- and low-power dual-core processors, delineates the unit responsibilities for battery monitoring, false voltage determination, and data decision-making, eliminates functional ambiguities, provides complete structural support, and has no logical loopholes. Second, it deeply integrates predictive information into each branch of hierarchical decision-making, enabling weight correction, threshold fine-tuning, and priority optimization, providing a basis for four-dimensional joint decision-making. Third, it eliminates the hidden danger of ineffective wake-up power consumption by adding low-power main control pre-judgment timing logic. After the cycle is reached, it first verifies the basic battery conditions before deciding whether to wake up the coprocessor unit, eliminating ineffective wake-up power consumption and aligning with the original intention of low-power design. Fourth, this invention uses absolute voltage threshold determination instead of capacity percentage recovery, making the low-voltage cache retransmission strategy physically feasible and avoiding the system entering a dead loop of "waiting for power recovery." The threshold for determining false voltage has been broadened to a reasonable engineering range, and an anti-jitter mechanism has been added. New mechanisms such as a battery undervoltage protection threshold, voltage hysteresis judgment, batch upload, and energy safety net constraints have been introduced to address engineering issues in existing technologies, such as crashes during forced low-voltage uploads, power consumption spikes during batch transmissions, power consumption oscillations due to false voltage, and rapid power loss due to low voltage. Intelligent decision-making with layered fault tolerance balances timeliness and battery life. For high-urgency events, timeliness and downtime prevention strategies are employed; for routine data, battery life and oscillation prevention strategies are implemented; and for abnormal data, energy safety net constraints are added, dynamically balancing monitoring accuracy and device battery life. The prediction mechanism is deeply integrated into the decision-making system, adding confidence level classification, timeout rollback, and misjudgment fault tolerance mechanisms. The sample library supports online updates, avoiding invalid power consumption and template aging issues. The dual-RF monitoring architecture and wake-up process have been optimized. In sleep mode, the low-power main core independently parses the wake-up command, resolving process interruption issues and balancing the flexibility of remote wake-up with the advantages of ultra-low power consumption and battery life. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall logical structure of the low-power wireless sensor network gateway of the present invention.

[0018] Figure 2 This is a schematic diagram of the internal functional architecture of the edge computing and power coprocessing unit of the present invention.

[0019] Figure 3 This is a schematic diagram of the four-dimensional hierarchical decision-making logic of the present invention.

[0020] Figure 4 This is a schematic diagram of the overall process of the gateway control method of the present invention.

[0021] Figure 5 This is a schematic diagram illustrating the detailed process of generating and executing the joint strategy of this invention.

[0022] Figure 6 This is a schematic diagram of the predictive power consumption optimization and decision fusion fault-tolerant process of the present invention.

[0023] Figure 7This is a schematic diagram of the low-power remote wake-up process of the present invention;

[0024] Figure 8a A timing diagram for fixed-period uploads using existing technology;

[0025] Figure 8b This is a schematic diagram of the low-urgency routine data upload timing of the present invention;

[0026] Figure 8c This is a schematic diagram of the high-urgency event upload timing of the present invention;

[0027] For example: 11-Wireless public network communication unit; 12-Data processing and control unit; 13-Low-power wake-up radio frequency unit; 14-Wireless personal area network radio frequency unit; 15-Battery; 16-Power management module; 17-Storage unit; 121-Low-power resident main control unit; 122-Edge computing and power co-processing unit; 1221-Edge computing module; 1222-Power status awareness module; 1223-Joint decision engine module; 1224-Policy output module; 1225-Time series learning and predictive optimization module. Detailed Implementation

[0028] 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.

[0029] like Figure 1 As shown, a low-power wireless sensor network gateway includes a wireless public network communication unit 11, a data processing control unit 12, a low-power wake-up radio frequency unit 13, a wireless personal area network radio frequency unit 14, a battery 15, and a power management module 16.

[0030] The wireless public network communication unit 11 is used to establish public network data communication with a remote server. In this embodiment, the public network is the Internet, and the wireless public network communication unit 11 adopts a Quectel EC20 series 4G communication module. The peak power consumption of the 4G communication module for data transmission and reception can reach over 300mA. The wireless public network communication unit 11 is in a completely power-off sleep state by default, and is only briefly turned on when data needs to be uploaded or a remote task issued by the server needs to be executed. The power supply is immediately cut off after the task is completed.

[0031] The wireless personal area network (PAN) radio frequency unit 14 is used to establish local data communication with the wireless sensor. The PAN radio frequency unit 14 uses an industrial-grade LoRa communication module, with a continuous listening power consumption of approximately 5mA. The PAN radio frequency unit 14 remains in a low-power listening state, continuously receiving monitoring data uploaded by the wireless sensor. The wireless sensor is a vibration-temperature composite sensor installed on the device under test, continuously collecting vibration and temperature signals from the device.

[0032] The static power consumption of the low-power wake-up radio unit 13 is lower than that of the wireless public network communication unit 11. The low-power wake-up radio unit 13 adopts a LoRa communication module of model Ra-01SC, and the continuous listening reception power consumption is about 5mA. The low-power wake-up radio unit 13 is in a low-power listening state and is used to receive remote wake-up commands sent by the remote server through the LoRa micro base station.

[0033] The data processing and control unit 12 adopts a high- and low-power dual-core collaborative architecture, including a low-power resident master control unit 121 and an edge computing and power co-processing unit 122. The low-power resident master control unit 121 uses an STMicroelectronics STM32L0 series ultra-low-power microcontroller, with a standby power consumption of less than 1μA. The low-power resident master control unit 121 operates continuously with power on, responsible for real-time monitoring of the battery 15's power status, monitoring battery voltage fluctuations, determining battery voltage dips, performing upload cycle timing, managing sleep mode, and parsing and verifying remote wake-up commands. The edge computing and power co-processing unit 122 uses an STMicroelectronics STM32H7 series high-performance microcontroller with a maximum clock frequency of 550MHz, possessing powerful digital signal processing capabilities. The edge computing and power co-processing unit 122 only wakes up to run on demand and is completely powered off after task completion, with no static power consumption.

[0034] A dedicated SPI or I2C data interaction interface is configured between the low-power resident master control unit 121 and the edge computing and power coprocessing unit 122. This interface is used to transmit the battery terminal voltage, voltage stability, virtual voltage state, and static recovery state power monitoring data cached in the low-power resident master control unit 121 during standby to the edge computing and power coprocessing unit 122 in batches after the edge computing and power coprocessing unit 122 is woken up and powered stably.

[0035] The edge computing and power coprocessing unit 122 is used to perform edge computing on the collected monitoring data to obtain feature data, determine the urgency of the data to be uploaded based on the feature data matching results, and receive battery power status data transmitted by the low-power resident main control unit 121, and generate a joint power management strategy in combination with the battery power status.

[0036] Furthermore, to prevent the I / O pins of the edge computing and power coprocessor unit 122 from forming a parasitic power supply loop with the powered low-power resident master unit 121 through the dedicated SPI / I2C data interface when the edge computing and power coprocessor unit 122 is completely powered off, thus avoiding additional leakage and unexpected startup, the power management module 16 of this gateway, while cutting off the power supply to the edge computing and power coprocessor unit 122, simultaneously cuts off the power supply to the interface level converter and pull-up resistor of the dedicated data interface through an independent load switch, so that the interface is in a physically high-impedance isolated state during the power-off period of the coprocessor unit. Power to the interface is restored and data communication is activated only after the low-power resident master unit 121 wakes up the coprocessor unit as needed and the power supply to the coprocessor unit is stably established (delay ≥ 5ms). This eliminates the bus leakage path in the sleep state and ensures that the static power consumption of the entire gateway strictly meets the design expectations.

[0037] The power management module 16 is electrically connected to the wireless public network communication unit 11, the data processing control unit 12, and the battery 15. The power management module 16 integrates multiple independent load switches and a high-precision power detection circuit, enabling it to execute a unified power management strategy and perform independent, precise power-on / off control on each functional unit. The battery 15 uses a 10000mAh capacity 18650 lithium battery pack with a nominal voltage of 3.7 volts, providing operating power to all gateway hardware units.

[0038] Storage unit 17 uses an SPI interface NOR Flash memory with a capacity of 128Mbit, which is used to store monitoring data, feature sample data, gateway firmware and configuration parameters.

[0039] like Figure 2 As shown, the edge computing and power coprocessing unit 122 integrates an edge computing module 1221, a power status awareness module 1222, a joint decision engine module 1223, and a policy output module 1224.

[0040] The edge computing module 1221 is used to extract and process features from the raw monitoring data to obtain feature data that characterizes the operating status of the device under test. In this embodiment, the raw vibration time-domain data collected by the wireless vibration sensor is the monitoring data. The edge computing module 1221 performs a fast Fourier transform on the data to calculate statistical features such as root mean square value, kurtosis, and skewness. The resulting frequency domain features or statistical features are the feature data.

[0041] Construction and online adaptive update mechanism of equipment operation feature sample library:

[0042] The device operation feature sample data pre-stored in storage unit 17 includes: the first sample data during the device startup phase (corresponding to a steep change in the rising edge of the vibration envelope), the second sample data during the device shutdown phase (corresponding to the vibration amplitude decaying to zero), and the third sample data during the device normal operation phase (corresponding to the feature quantity being within the steady-state threshold band). The initial template of the above sample data is obtained during the gateway deployment and debugging phase in the following manner: under the known normal operating conditions of the device under test, vibration / temperature monitoring data for no less than 10 complete operating cycles are continuously collected; the edge computing module 1221 extracts the same feature vector (root mean square value, kurtosis, frequency domain main peak amplitude, etc.) for each cycle, uses K-means clustering to remove outlier cycles, takes the arithmetic mean of the feature vectors of the remaining cycles, generates the initial sample template, and stores it in NOR Flash storage unit 17.

[0043] In actual long-term operation, the sample library has the ability to update online adaptively. Whenever the gateway confirms through the remote server or determines through local preset rules that the current device status is correct and the operation is stable, the edge computing module 1221 automatically updates the current real-time feature vector with the historical template by weighted moving average: new template = 0.9 × old template + 0.1 × current feature vector (weights are configurable). This allows the sample library to follow the gradual changes in characteristics such as mechanical wear and tear of the device and drift in operating conditions, avoiding mismatches caused by the aging of fixed templates.

[0044] The matching determination quantization threshold uses a normalized Euclidean distance metric in the feature data matching process. Specifically, the Euclidean distance between the real-time feature vector and the three types of sample templates is calculated and normalized by dividing by the standard deviation of the feature dimension. When the normalized distance is less than the preset matching threshold (set to 0.82 in this embodiment), it is determined to be a match; if the distance with all sample templates is greater than the threshold, it is determined to be an abnormal device operating status.

[0045] The power status sensing module 1222 works in conjunction with the low-power resident master control unit 121 to receive and analyze information such as remaining battery power, battery terminal voltage, instantaneous power consumption, and battery voltage fluctuation and stability status. In this embodiment, the low-power resident master control unit 121 continuously pushes battery status data to the power status sensing module 1222 via an SPI or I2C interface, providing accurate energy status input for joint decision-making.

[0046] The joint decision engine module 1223 incorporates a multi-dimensional joint decision model based on data urgency and battery status. This model integrates four dimensions—data service urgency, remaining battery power, battery physical stability, and device operation prediction information—to generate optimal power consumption control and data upload strategies. Device operation prediction information is used to correct data urgency weights, dynamically fine-tune voltage judgment thresholds, and participate in the entire hierarchical fault-tolerant decision-making process. This is the core intelligent component of the invention, integrating data value and energy status analysis to eliminate the drawbacks of data processing and power management being separated in traditional solutions. In the multi-dimensional joint decision model, when the remaining power percentage reported by the low-power resident main control unit 121 conflicts with the energy status indicated by the measured terminal voltage, the measured terminal voltage result is used as the priority for decision-making.

[0047] The strategy output module 1224 is used to convert the joint strategy generated by the decision model into hardware control instructions, which are then sent to the power management module 16 for execution. The strategy instructions include specific start-up timing, strategy upload timing, and shutdown timing, and achieve precise power-on and power-off control of each functional unit by controlling the independent load switches in the power management module 16.

[0048] like Figure 3 As shown, the joint decision engine module 1223 is configured with a hierarchical fault-tolerant decision mechanism, which specifically includes a high-urgency data processing strategy, a low-urgency data processing strategy, and an abnormal data dynamic adjustment strategy.

[0049] For high-urgency event data corresponding to device startup and shutdown, a battery voltage undervoltage protection threshold is set. In this embodiment, the undervoltage protection threshold is set to 3.3V (corresponding to approximately 5%~8% of no-load SOC). When the feature data matches the pre-stored first sample data, it is determined to be a device startup event; when the feature data matches the pre-stored second sample data, it is determined to be a device shutdown event. When the battery voltage is higher than the undervoltage protection threshold of 3.3V, the wireless public network communication unit 11 is immediately triggered to upload data, and power is cut off immediately after the upload is completed. When the battery voltage is lower than the undervoltage protection threshold of 3.3V, high-priority data is encrypted and cached to the storage unit 17. Data is re-uploaded after the open-circuit voltage caused by the battery polarization voltage decay has been allowed to recover to the preset safe startup voltage threshold (set to 3.6V in this embodiment) and the voltage fluctuation has stabilized. This recovery process is an electrochemical polarization recovery phenomenon and does not represent an increase in the percentage of remaining battery capacity. The re-upload condition is based on the absolute voltage threshold. The principle behind this mechanism is as follows: when the battery voltage is extremely low, the instantaneous high current when the 4G communication module is turned on may further pull the battery voltage below the chip reset threshold, causing the gateway to crash and restart and data loss. By setting an undervoltage protection threshold and a buffer-based transmission mechanism based on absolute voltage recovery, the system instability risk caused by low-voltage high-current surges is avoided, while ensuring the eventual availability of critical data. When the device operation prediction information anticipates an impending start-up or shutdown event, it proactively increases the urgency weight of the corresponding event data and prioritizes reserving power resources.

[0050] For routine monitoring data with low urgency corresponding to normal device operation, a terminal voltage safety threshold and hysteresis judgment mechanism are configured. In this embodiment, the safety threshold is set to 3.5V, and the hysteresis range is set to 3.48V to 3.52V. Battery voltage fluctuations and stable voltage status are continuously monitored and judged by the low-power resident main control unit 121 and pushed in real time. When the feature data matches the pre-stored third sample data, it is judged as routine monitoring data. Batch uploads are only performed when the preset upload cycle is reached, the battery terminal voltage is higher than the safety threshold, and the battery voltage status is stable without voltage fluctuations or significant fluctuations. If the voltage is in the critical range of 3.48V to 3.52V, the hysteresis anti-jitter mechanism is activated to delay the upload judgment and avoid frequent start-stop power consumption oscillations caused by battery voltage fluctuations. If the voltage is lower than the safety threshold for a long period of time, the data continues to be temporarily stored and low-power sleep mode is maintained. If it is predicted that the device will be in a long-term stable state, the periodic upload threshold is appropriately relaxed to further reduce the wake-up frequency.

[0051] For high-urgency anomaly data corresponding to equipment malfunctions, a mismatch between the feature data and the third sample data is identified as an equipment anomaly. After real-time uploading, the sensor data acquisition frequency and edge computing execution frequency are dynamically adjusted based on the degree of anomaly deviation, and an energy safety net constraint is set in conjunction with the battery voltage status. In this embodiment, if the vibration kurtosis value exceeds the normal upper limit by 10%, the acquisition and calculation frequency is slightly increased to once per minute; if it exceeds the normal upper limit by more than 50%, the frequency is significantly increased to once every 10 seconds. When the battery voltage is low, for example below 3.6V, the frequency modulation amplitude is converged, with an upper limit limited to once per minute, balancing monitoring accuracy and equipment battery life. When predicting the evolution trend of equipment failure, the priority of anomaly monitoring is increased in advance.

[0052] Battery voltage fluctuation and stability assessment criteria (adapted to industrial electromagnetic environments). Considering the DC-DC power supply ripple inside the wireless gateway, LoRa RF spurious emissions, and millivolt-level noise introduced by PCB layout, the voltage stability assessment threshold is set within a reasonable engineering range. The specific criteria are: continuously monitoring the difference between the peak and peak battery voltage fluctuations within a 100ms sliding time window. If the fluctuation difference is less than or equal to 50mV (this value is much greater than the typical PCB ripple noise floor and sufficient to suppress false judgments caused by voltage fluctuations), the voltage is considered stable and there is no voltage fluctuation; if the fluctuation difference is greater than 50mV, the battery voltage is considered to be fluctuating drastically or in a voltage fluctuation state, and low-urgency routine data uploads are prohibited. A jitter-reducing hysteresis mechanism is also introduced: only after three consecutive samples (each 100ms apart) meet the stability condition is the state considered validly stable, preventing frequent state reversals caused by sudden interference in a single sample.

[0053] like Figure 4 As shown, the method specifically includes the following steps.

[0054] Step S1: The gateway is in standby mode. The low-power resident main control unit 121 monitors the battery status, battery voltage stability and false voltage status in real time, and executes the running cycle timing. The wireless personal area network radio frequency unit 14 continuously collects sensor monitoring data, and the low-power wake-up radio frequency unit 13 continuously listens for remote wake-up commands.

[0055] Step S2: Upon acquiring sensor monitoring data or reaching the preset upload cycle, the low-power resident main control unit 121 prioritizes completing the basic terminal voltage and voltage status pre-judgment. Only when the basic wake-up conditions are met is the edge computing and power co-processing unit 122 woke up as needed to perform edge computing feature extraction of monitoring data and accurate verification of battery power status in parallel. This step reflects the key timing optimization of the present invention: when the cycle arrives, the low-power resident main control unit 121 first autonomously completes the pre-judgment of voltage threshold, voltage stability, and no false voltage. Only when all pre-judgment conditions are met is the edge computing and power co-processing unit 122 woke up, eliminating invalid wake-up power consumption.

[0056] Step S3: Match the extracted feature data with pre-stored device operation feature sample data to determine the urgency of the current data. Combine the four-dimensional parameters of battery terminal voltage, battery stability, and device operation prediction information to generate a multi-dimensional joint power management and data upload strategy. The feature sample data includes the first sample data during the device under test's startup phase, the second sample data during the device under test's shutdown phase, and the third sample data during the device under test's normal operation. The matching process is achieved by calculating the normalized Euclidean distance between the feature data and each sample data. A match is considered complete when the similarity exceeds a preset threshold (0.82 in this embodiment).

[0057] Step S4: Execute the corresponding control strategy according to the hierarchical fault-tolerant decision-making mechanism. High-urgency data is uploaded on demand or cached and re-uploaded with priority. Low-urgency data is uploaded in batches at different times. Abnormal data is dynamically adjusted in terms of collection and processing frequency and energy fallback constraints are applied. The prediction results are dynamically adjusted to correct the decision weights, so as to achieve deep integration of prediction and hierarchical decision-making.

[0058] Step S5: After all data processing, uploading, and device interaction tasks are completed, immediately cut off the power supply to the edge computing and power co-processing unit 122 and the wireless public network communication unit 11, leaving only the low-power resident main control unit 121, the low-power wake-up radio unit 13, and the wireless personal area network radio unit 14 in standby mode, and iteratively execute the monitoring and control process. Specifically, the low-power resident main control unit 121 maintains microampere-level standby, while the low-power wake-up radio unit 13 and the wireless personal area network radio unit 14 maintain milliampere-level low-power monitoring.

[0059] like Figure 5 As shown, the process further refines and expands steps S3 and S4.

[0060] When the feature data matches the device startup or shutdown sample, it is determined to be a high-urgency event, and the terminal voltage limit threshold protection strategy is executed. If the battery terminal voltage is higher than 3.3V, the wireless public network communication unit 11 is immediately activated to upload data, and the power is cut off immediately after the upload is completed. If the battery terminal voltage is lower than 3.3V, the event data is encrypted, marked as high priority, and cached in the storage unit 17. It is re-uploaded after the battery terminal voltage recovers to 3.6V and stabilizes. Combining the device's predicted start-up and shutdown trend, the event priority is increased in advance, and power consumption resources are reserved in advance. This mechanism avoids device downtime and data loss caused by low-voltage high-current surges. For the recovery threshold (safe startup voltage), the embodiment was initially set to 3.4V (only 0.1V higher than the undervoltage protection threshold). However, during prototype testing, it was found that the voltage recovery during battery polarization decay exhibits an exponential decay characteristic. While it only takes about 5-10 seconds for the voltage to recover from 3.3V to 3.4V, the instantaneous voltage drop during 4G module startup under load pulls the voltage down below 3.3V again, creating an oscillating state of "undervoltage immediately upon startup," leading to repeated data retransmission failures. Subsequently, an attempt was made to increase the recovery threshold to 3.7V, but actual testing revealed that this voltage recovery required a wait of 3-15 minutes (depending on the depth of discharge and battery temperature), resulting in excessively long delays for critical data and violating the timeliness requirements for high-urgency events. Ultimately, a compromise was made with 3.6V—this value leaves a 0.3V voltage drop margin to withstand the impact of 4G module startup, while the typical waiting time is controlled between 30 seconds and 3 minutes, achieving a balance between reliability and timeliness.

[0061] When the feature data matches samples from normal device operation, it is determined to be low-urgency routine data. The monitoring data is temporarily stored in storage unit 17, combined with end-voltage hysteresis judgment, battery voltage stability judgment, and a batch upload mechanism. The low-power resident main control unit 121 keeps a real-time timer. Upon reaching the preset upload cycle, it autonomously completes pre-judgments of end-voltage threshold, voltage stability, and absence of voltage hysteresis. Only when all pre-judgment conditions are met is the edge computing and power co-processing unit 122 awakened to perform batch uploads. If the pre-judgment is not met, the wake-up process is skipped, and the system remains in sleep mode. This mechanism effectively suppresses power consumption fluctuations caused by battery voltage hysteresis, preventing high-power batch transmissions from depleting battery power. The upload cycle threshold is dynamically optimized based on steady-state operation prediction results, further reducing the wake-up frequency.

[0062] When the feature data does not match the normal operating sample, it is determined to be an equipment malfunction. The malfunction data is uploaded immediately, and the monitoring frequency is dynamically adjusted according to the malfunction level and the remaining battery level.

[0063] Battery resting recovery quantification criteria (based on the principle of electrochemical polarization recovery). When the battery terminal voltage falls below the preset undervoltage protection threshold V_low (e.g., 3.3V, corresponding to approximately 5%~8% of the no-load SOC) and high-urgency data is cached, the low-power resident main control unit 121 continuously monitors the battery open-circuit terminal voltage. It should be noted that: high-current discharge (such as a sudden transmission from a 4G communication module) will cause an increase in the internal polarization resistance of the battery, resulting in a momentary drop in terminal voltage; when the high-power load is disconnected, the polarization voltage will gradually subside, and the terminal voltage will exhibit a physical recovery phenomenon. This recovery is an electrochemical reversible recovery process and does not represent a real increase in the battery's remaining capacity (SOC, percentage), but only a safe window for the battery to have short-term load-carrying capacity.

[0064] Based on the above principles, the retransmission trigger condition is set to an absolute voltage threshold judgment, rather than a percentage recovery judgment: when the battery terminal voltage steadily recovers to the preset safe start voltage V_start (e.g., 3.6V) for 3 consecutive seconds, and the voltage fluctuation range within this 3-second window is ≤50mV, it is determined that the static recovery is complete and the power consumption start condition is met, triggering the edge computing and power coprocessing unit 122 to wake up and perform cached data retransmission. If the terminal voltage drops below V_low again during subsequent retransmission, the upload is immediately stopped and the static monitoring state is re-entered to avoid system crashes due to undervoltage reset. Combined with abnormal trend prediction, the monitoring strategy is strengthened in advance to achieve a dynamic balance between monitoring accuracy and battery life. When the remaining power percentage (SOC) reported by the battery management system (BMS) conflicts with the energy state indicated by the measured terminal voltage (e.g., the SOC is high but the terminal voltage is close to the undervoltage threshold), the measured terminal voltage result shall prevail, because the terminal voltage is directly related to the hardware undervoltage reset risk, ensuring that the gateway can still maintain basic decision-making survivability under the worst physical conditions. Meanwhile, the aforementioned recovery waiting time window is not a fixed value and changes dynamically according to battery temperature, aging degree and polarization decay rate. The typical window range is 3 seconds to 5 minutes. The low-power resident main control unit continuously monitors during this period until the terminal voltage reaches the standard or the battery is depleted and enters the final protection state. During this period, no high-power wake-up operation is triggered to avoid invalid waiting power consumption.

[0065] like Figure 6 As shown, the timing learning and predictive optimization module 1225 built into the edge computing and power coprocessing unit 122 performs the following process.

[0066] In daily operation, the edge computing and power coprocessing unit 122 continuously records the time-series historical data of the device under test during startup, shutdown, and abnormal operation, and stores each state change event and its timestamp in the storage unit 17.

[0067] After accumulating sufficient historical data, such as a month of operation records, the time-series learning and predictive optimization module 1225 performs learning and analysis on the stored operation event data. In this embodiment, a time-series pattern mining algorithm is used to statistically analyze the frequency of start-up and shutdown events on each day of the week and at each time period of the day. For example, the learning may discover that a certain device starts up at 8:00 AM every Monday to Friday, shuts down at 8:00 PM every evening, and is permanently shut down on Saturdays and Sundays.

[0068] Based on historical operating patterns, future device status change nodes are predicted, and pre-startup preparation operations are performed according to confidence level classification. Specifically, the confidence level classification mechanism is as follows: when the prediction confidence level is below 80%, only algorithm parameters are preloaded to storage unit 17, and no high-power hardware modules are activated. When the prediction confidence level is above 80%, the communication link parameters of the edge computing and power co-processing unit 122 are pre-initialized 1 to 3 seconds in advance, but the wireless public network communication unit 11 is not activated.

[0069] If the predicted timeout fails to trigger the corresponding device event, the system actively reverts to deep sleep mode. A 5-second prediction window is set. If the wireless personal area network radio frequency unit 14 fails to collect the corresponding device status data within the window, the preloaded resources are immediately cleared, and the system returns to deep sleep mode, eliminating unnecessary power consumption caused by prediction errors. Simultaneously, the feature sample library is updated online using a weighted moving average mechanism to avoid misjudgments caused by aging fixed feature templates.

[0070] like Figure 7 As shown, the gateway enters a deep sleep state when there are no tasks. At this time, the wireless public network communication unit 11 is completely powered off, and the edge computing and power co-processing unit 122 is also completely powered off. The low-power resident main control unit 121 maintains a microampere-level (<1μA) standby state, while the low-power wake-up RF unit 13 and the wireless personal area network RF unit 14, due to their RF physical layer characteristics, maintain a milliampere-level low-power listening state (typically about 5mA). The total static current of the entire gateway in this deep sleep mode is typically about 5.1mA, which falls within the ultra-low power industrial grade category.

[0071] When the remote server needs to query gateway data or issue firmware upgrade commands via the public network, the server issues a remote wake-up command through the LoRa micro base station deployed on-site. After receiving the wake-up command that matches the address of this gateway, the low-power wake-up radio unit 13 forwards the command data to the low-power resident master control unit 121.

[0072] The low-power resident master control unit 121 independently completes instruction parsing and validity verification, including address matching, integrity verification, and permission verification. Since the low-power resident master control unit 121 operates with power on for extended periods, the entire process of receiving and parsing wake-up instructions does not rely on the power-off edge computing and power coprocessing unit 122, thus solving the logical break problem in the background technical solution where the core unit cannot parse remote instructions after power failure.

[0073] After successful verification, the low-power resident main control unit 121 wakes up the edge computing and power coprocessing unit 122 and the wireless public network communication unit 11 as needed. Once activated, the edge computing and power coprocessing unit 122 controls the wireless public network communication unit 11 to establish a public network connection with the server, based on the task type in the wake-up command, and performs tasks such as remote data querying, firmware upgrades, and parameter configuration. After all remote tasks are completed, the wireless public network communication unit 11 and the edge computing and power coprocessing unit 122 are immediately shut down, returning to deep sleep mode to minimize power consumption in remote operation and maintenance scenarios.

[0074] Figure 8a The typical timing of the background technical solution is illustrated. During normal device operation, this solution uploads data at fixed intervals, activating the public network communication unit each time, generating a power consumption spike. Even as the battery voltage gradually decreases to a lower level, the solution continues to upload data at the fixed intervals, with each upload's power consumption spike further consuming the already insufficient power. When the battery voltage drops to an extremely low level, the instantaneous high current from activating the public network module may momentarily pull the battery voltage below the chip reset threshold, causing the gateway to crash or experience irreversible power loss.

[0075] Figure 8b This paper demonstrates the timing of the present invention's solution in a low-urgency, routine monitoring data scenario. The present invention also uploads data according to a preset cycle, but the key difference lies in that at each cycle point, the low-power resident main control unit 121 autonomously performs a pre-judgment of the battery status. If, at time point T1, the battery voltage is detected to be below the safety threshold or there is a false voltage, the wake-up process is skipped, and data is temporarily stored, with the gateway remaining in deep sleep mode. During the time period from T1 to T2, the gateway power consumption remains at the basic microamp to milliamp level. Only when the battery voltage recovers to above the safety threshold and is stable without false voltage at time T3, is the coprocessor unit woken up as needed to perform batch uploads. This strategy effectively avoids consuming valuable energy by performing non-urgent tasks under low voltage conditions, ensuring the continuous survival of the gateway.

[0076] Figure 8cThis demonstrates the timing of the present invention's solution in a high-urgency event scenario. When a high-urgency event such as device startup, shutdown, or abnormality is detected at time point T4, if the battery voltage is higher than the undervoltage protection threshold of 3.3V, the joint decision engine module 1223 generates an immediate upload strategy, ignoring conventional power constraints, and immediately activates the public network communication unit to execute the upload task. If the battery voltage is lower than the undervoltage protection threshold of 3.3V, a strategy of encrypted caching and waiting for the battery voltage to recover to 3.6V due to polarization decay before re-uploading is implemented to avoid system crashes caused by low-voltage surges. After the upload is completed, the system immediately returns to the shutdown state, ensuring the timeliness of critical event information.

[0077] Through the detailed description of the above embodiments, this invention utilizes a high- and low-power dual-core collaborative architecture, a four-dimensional hierarchical fault-tolerant joint decision-making mechanism, a static recovery judgment based on an absolute voltage threshold, an engineering-adapted virtual voltage anti-jitter mechanism, a complete sample library construction and online update method, and a graded radio frequency wake-up technology solution. The gateway's average power consumption drops to an extremely low level (approximately 5.1mA total static) during most non-urgent periods, while ensuring real-time reporting capabilities for critical events and flexibility in remote management, making it suitable for unattended industrial monitoring scenarios requiring long-term battery power. This invention employs an intelligent decision-making process that assesses urgency, combines power consumption, and dynamically generates strategies. This method simultaneously handles two tasks: "data upload" and "power management." The data and power balance logic avoids the risk of system downtime due to uploading non-urgent data under low power conditions, reflecting an intelligent trade-off for device lifecycle. For high-urgency events such as device start-up / shutdown and anomalies, the decision logic actively ignores power limitations and executes "immediate upload." This ensures the absolute timeliness of critical information and achieves optimal task priority scheduling under power budget constraints. A learning mechanism is introduced, enabling the gateway to predict device status. This predictive capability in the time dimension works in conjunction with power management to achieve low-latency control from passive response to proactive preparation through a "pre-start strategy," further optimizing the timeliness of critical tasks.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A low-power wireless sensor network gateway, characterized in that, It includes a wireless public network communication unit (11), a data processing control unit (12), a low-power wake-up radio frequency unit (13), a wireless personal area network radio frequency unit (14), a battery (15), and a power management module (16). The wireless public network communication unit (11) is used to establish public network data communication with the remote server and is in a completely power-off sleep state by default. The wireless personal area network radio frequency unit (14) is used to establish local data communication with the wireless sensor, and is continuously in a low-power listening state to collect sensor monitoring data. The static power consumption of the low-power wake-up radio frequency unit (13) is lower than that of the wireless public network communication unit (11), and it is in a low-power listening state to receive remote wake-up commands sent by the remote server. The data processing control unit (12) includes a low-power resident main control unit (121) and an edge computing and power coprocessing unit (122). The low-power resident main control unit (121) is powered on and in standby mode for a long time. It is used to monitor the battery power status, battery voltage fluctuation status, execution cycle timing, hibernation management, and wake-up command parsing in real time. The edge computing and power coprocessing unit (122) is a high-performance computing unit that is only woken up and run on demand. It is completely powered off after the task is completed. A dedicated data interaction interface is set between the low-power resident main control unit (121) and the edge computing and power coprocessing unit (122). After the edge computing and power coprocessing unit (122) is woken up and powered on stably, it transmits the battery voltage, voltage stability, false voltage status, and static recovery status power monitoring data cached in the low-power resident main control unit (121) during standby to the edge computing and power coprocessing unit (122) in batches. The edge computing and power coprocessing unit (122) is used to perform edge computing on the collected monitoring data to obtain feature data, determine the urgency of the data to be uploaded based on the feature data matching result, and receive the battery power status data transmitted by the low-power resident main control unit (121), and generate a joint power management strategy in combination with the battery power status. The power management module (16) is electrically connected to the wireless public network communication unit (11), the data processing control unit (12), and the battery (15) respectively, and is used to execute the joint power management strategy and perform fine-grained power-on and power-off control on each functional unit. The battery (15) is used to provide operating power to the entire gateway hardware unit.

2. The low-power wireless sensor network gateway according to claim 1, characterized in that, The edge computing and power coprocessing unit (122) includes an edge computing module (1221), a power status awareness module (1222), a joint decision engine module (1223), and a strategy output module (1224). The edge computing module (1221) is used to extract and process features from the raw monitoring data to obtain feature data that characterizes the operating status of the device under test. The power status sensing module (1222) is used in conjunction with the low-power resident main control unit (121) to receive and parse the remaining battery power, battery terminal voltage, instantaneous power consumption, and battery false voltage and voltage stability information. The joint decision engine module (1223) has a built-in multi-dimensional joint decision model for data urgency and battery status. It combines four-dimensional parameters, including data service urgency, remaining battery power, battery physical stability, and device operation prediction information, to generate the optimal power consumption control and data upload strategy. The device operation prediction information is used to correct the data urgency weight, dynamically fine-tune the voltage judgment threshold, and participate in the whole process of hierarchical fault-tolerant decision-making. In the multi-dimensional joint decision-making model, when the remaining power percentage reported by the low-power resident main control unit (121) conflicts with the energy state indicated by the measured terminal voltage, the measured terminal voltage judgment result is used as the decision priority basis. The strategy output module (1224) is used to convert the joint strategy generated by the decision model into hardware control instructions and send them to the power management module (16) for execution.

3. The low-power wireless sensor network gateway according to claim 2, characterized in that, The joint decision engine module (1223) is configured with a hierarchical fault-tolerant decision mechanism, which specifically includes a high-urgency data processing strategy, a low-urgency data processing strategy, and an abnormal data dynamic adjustment strategy. Set a battery voltage undervoltage protection threshold for high-urgency event data corresponding to equipment startup and shutdown; When the battery terminal voltage is higher than the undervoltage protection threshold, the wireless public network communication unit (11) is immediately triggered to upload data, and the power is cut off immediately after the upload is completed; when the battery terminal voltage is lower than the undervoltage protection threshold, high priority data is encrypted and cached, and the data is re-uploaded after the open-circuit terminal voltage caused by the decline of polarization voltage is allowed to rise to the preset safe start-up voltage threshold and the voltage fluctuation is stable; this rise process is an electrochemical polarization recovery phenomenon, which does not represent an increase in the percentage of the remaining battery capacity, and the re-upload condition is based on the absolute voltage threshold; when the equipment operation prediction information predicts that a start-up and shutdown event is about to occur, the urgency weight of the corresponding event data is actively increased, and power consumption resources are reserved in priority. For routine monitoring data with low urgency corresponding to normal equipment operation, a terminal voltage safety threshold and hysteresis judgment mechanism are configured. Battery false voltage and voltage stability are continuously monitored and judged by the low-power resident main control unit (121) and pushed in real time. Only when the preset upload cycle is reached, the battery terminal voltage is higher than the safety threshold and the battery voltage is stable without false voltage or large fluctuations, batch upload is performed. At other times, the data is temporarily stored and low-power sleep is maintained. If it is anticipated that the device will be in a long-term steady-state operation, the periodic upload threshold should be appropriately relaxed to further reduce the wake-up frequency; For high-urgency abnormal data corresponding to equipment malfunctions, after real-time uploading, the sensor data acquisition frequency and edge computing execution frequency are dynamically adjusted according to the degree of abnormal deviation. Energy safety constraints are set in conjunction with the battery voltage status. When the voltage is low, the frequency modulation amplitude is reduced to balance monitoring accuracy and equipment battery life. When predicting the evolution trend of equipment failure, the priority of abnormal monitoring is increased in advance.

4. The low-power wireless sensor network gateway according to claim 1, characterized in that, The edge computing and power coprocessing unit (122) also has a built-in timing learning and predictive optimization module (1225). The time-series learning and predictive optimization module (1225) is used to continuously record the time-series historical data of the start-up, shutdown, and abnormal operation of the tested equipment. It learns the periodic operation mode of the equipment through time-series mining, generates equipment operation prediction information, and inputs it into the joint decision engine module (1223) in real time to participate in multi-dimensional joint decision-making. The time-series learning and predictive optimization module (1225) and the edge computing module (1221) share the feature sample library stored in the storage unit (17). The sample library is updated online on the edge computing module (1221) side according to the weighted moving average mechanism to avoid misjudgment caused by the aging of fixed feature templates. The timing learning and predictive optimization module (1225) is configured with a prediction confidence level mechanism and a timeout fault tolerance rollback mechanism. For high-urgency events, low-confidence predictions only preload algorithm parameters, while high-confidence predictions pre-wake up hardware units in a tiered manner. If no corresponding device event is detected within the prediction window, the preloaded resources are immediately cleared and deep sleep is restored to avoid ineffective power consumption.

5. The low-power wireless sensor network gateway according to claim 1, characterized in that, When the gateway is in deep sleep mode, the low-power resident main control unit (121) maintains microampere-level standby, and the low-power wake-up radio unit (13) and the wireless personal area network radio unit (14) maintain milliampere-level low-power listening mode. After receiving the remote wake-up command, the low-power wake-up radio frequency unit (13) independently completes the command parsing and legality verification by the low-power resident main control unit (121). After the verification is passed, the edge computing and power coprocessing unit (122) and the wireless public network communication unit (11) are woken up as needed to perform remote data interaction or firmware upgrade tasks. After the task is completed, the high-power unit is immediately shut down.

6. A control method for a low-power wireless sensor network gateway, characterized in that, The application of the wireless sensor network gateway according to any one of claims 1 to 5 includes the following steps: S1. The gateway is in standby mode. The low-power resident main control unit (121) monitors the battery status, battery voltage stability and false voltage status in real time and performs the running cycle timing. The wireless personal area network radio frequency unit (14) continuously collects sensor monitoring data. The low-power wake-up radio frequency unit (13) continuously listens for remote wake-up commands. S2. When sensor monitoring data is acquired or the preset upload period is reached, the low-power resident main control unit (121) prioritizes the basic end voltage and voltage status pre-judgment. Only when the basic wake-up conditions are met, the edge computing and power co-processing unit (122) is woken up as needed to complete the edge computing feature extraction of monitoring data and accurate verification of battery power status in parallel. S3. Match the extracted feature data with the pre-stored equipment operation feature sample data, determine the urgency of the current data, and generate a multi-dimensional joint power management and data upload strategy by combining four-dimensional parameters such as battery terminal voltage, battery stability, and equipment operation prediction information. S4. Execute corresponding control strategies according to the hierarchical fault-tolerant decision-making mechanism. High-urgency data is uploaded on demand or cached and re-uploaded with priority. Low-urgency data is uploaded in batches at different times. Abnormal data is dynamically adjusted in terms of collection and processing frequency and energy fallback constraints are applied. The prediction results are dynamically corrected to adjust the decision weights, thereby achieving deep integration of prediction and hierarchical decision-making. S5. After all data processing, uploading, and device interaction tasks are completed, immediately cut off the power supply to the edge computing and power coprocessing unit (122) and the wireless public network communication unit (11), and keep only the low-power unit in standby mode to iteratively execute the monitoring and control process.

7. The control method for a low-power wireless sensor network gateway according to claim 6, characterized in that, The joint policy generation process in step S3 specifically includes: When the feature data matches the device startup or shutdown sample, it is determined to be a high-urgency event, and the terminal voltage limit protection strategy is executed to avoid device downtime and data loss caused by low voltage and high current impact; combined with the device's predicted start-up and shutdown trend, the event priority is increased in advance; When the feature data matches the normal operating sample of the device, it is determined to be low-urgency routine data. Combined with the terminal voltage hysteresis judgment, battery false voltage stability judgment and batch upload mechanism, the power consumption oscillation caused by battery false voltage is suppressed, and the battery power is not depleted by batch high power transmission. The upload cycle threshold is dynamically optimized based on the steady-state operation prediction results. When the feature data does not match the normal operating sample, it is determined to be an abnormal event of the equipment. Abnormal data is uploaded in time, and the monitoring frequency is dynamically adjusted according to the abnormality level and the remaining battery level to achieve a dynamic balance between monitoring accuracy and battery life. Combined with the prediction of abnormal trends, the monitoring strategy is strengthened in advance.

8. The control method for a low-power wireless sensor network gateway according to claim 6, characterized in that, It also includes predictive low-power optimization steps: Continuously accumulate equipment runtime sequence data to uncover the temporal patterns of fixed start-up / shutdown and abnormal occurrences; predict future equipment state change nodes based on historical operating patterns, execute pre-start preparation operations based on confidence level classification, and input the prediction results into a hierarchical decision-making mechanism in real time to correct decision weights and thresholds; when the prediction timeout fails to trigger the corresponding equipment event, actively roll back to a deep sleep state to eliminate the unnecessary power consumption caused by prediction errors; the feature sample library is updated online according to a weighted moving average mechanism to avoid misjudgments caused by the aging of fixed feature templates.

9. The control method for a low-power wireless sensor network gateway according to claim 6, characterized in that, It also includes a remote low-power wake-up step: In the deep sleep state of the gateway, only the low-power resident master control unit (121) maintains microampere-level standby, while the low-power wake-up radio frequency unit (13) and the wireless personal area network radio frequency unit (14) maintain milliampere-level low-power listening state; after receiving the server wake-up command, the low-power resident master control unit (121) completes the command parsing and verification, wakes up the high-power unit as needed to execute the remote task, and immediately restores the sleep state after the task is completed, so as to minimize the overall power consumption of the remote wake-up scenario.

Images