Power supply intelligent protection control method and related intelligent power supply

Abstract

The invention discloses a power supply intelligent protection control method and a related intelligent power supply, and the method is beneficial to solving the problems of one-sided anomaly detection, rigid protection strategy, insufficient system collaboration and the like in traditional power supply protection through a technical path of multi-dimensional perception, intelligent model decision and hierarchical collaborative execution. And while the protection reliability is improved, the interference on load operation is reduced to the maximum extent, and the method has remarkable practicability and advancement.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power supply protection, in particular to a power supply intelligent protection control method and an intelligent power supply. BACKGROUND

[0002] An intelligent power supply is a power supply device with remote control and management functions. It can be connected through a network to realize remote monitoring, control and intelligent management of the power supply. Users can control the power supply switch of the device on any networked computer through the Internet or local area network, query, connect, disconnect or restart the power supply, without being limited by specific devices or programs. At the same time, the intelligent power supply can also manage and control multiple devices uniformly, improving the convenience and efficiency of power supply use.

[0003] The intelligent power supply has various practical functions. For example, it can automatically perform power on / off operations according to a preset period, which helps to save energy; it can monitor power supply voltage, current and other parameters to ensure the safe and stable operation of the device, and can also monitor the running state of the device to discover and handle power supply failures in a timely manner, improving the reliability of the device. In addition, the intelligent power supply usually has good scalability, and new functions or devices can be added according to needs.

[0004] The protection mechanism of the intelligent intelligent power supply in the prior art still has room for optimization. SUMMARY

[0005] In order to solve the above problems existing in the prior art, the embodiments of the present application provide a power supply intelligent protection control method and related intelligent power supply, which are beneficial to protect the safety of the intelligent power supply.

[0006] The first aspect provides an intelligent power supply protection method applied to a controller of an intelligent power supply, the intelligent power supply being used to supply power to a load device, the intelligent power supply further comprising a multi-dimensional sensor group, a communication module and an execution protection module, the multi-dimensional sensor group comprising a temperature sensor, a humidity sensor, a voltage sensor, a current sensor and a load state sensor, characterized in that the method comprises: acquiring multi-dimensional real-time data in a preset period through the multi-dimensional sensor group, the multi-dimensional real-time data comprising real-time voltage data, real-time current data, internal environment temperature data of the power supply, external environment humidity data of the power supply and real-time running state data of the load device at the power output end; normalizing the multi-dimensional real-time data based on a preset weight allocation rule to obtain a standardized data matrix, the weight allocation rule being set according to the type and power supply priority of the load device; inputting the standardized data matrix into a pre-trained anomaly detection model to output a real-time anomaly probability value, the anomaly detection model being a deep learning model based on a fusion of a long short-term memory network and an attention mechanism and being trained by historical fault sample data and normal operation sample data; if the real-time anomaly probability value is greater than a first preset threshold, acquiring an emergency level signal fed back by the load device through the communication module, the emergency level signal being generated by the load device according to a core function operation state of the load device; determining a protection level according to the real-time anomaly probability value and the emergency level signal through a preset decision tree model, the protection level including a warning level, a current limiting level, a short-time power-off level and a complete power-off level; controlling the execution protection module to perform a protection operation corresponding to the protection level and sending prompt information including an anomaly type and a protection measure to an associated terminal through the communication module.

[0007] Optionally, the real-time operation state data of the load device includes a CPU occupancy rate, a memory usage rate, a core component temperature and a priority identifier of a current running task of the load device.

[0008] Optionally, the normalization processing includes: mapping each dimension real-time data to an interval [0, 1], wherein a mapping reference of voltage data is a rated voltage range of the load device, a mapping reference of current data is a rated current range of the load device, a mapping reference of temperature data is a safe working temperature interval of the power supply, and a mapping reference of humidity data is a safe working humidity interval of the power supply.

[0009] Optionally, the training process of the anomaly detection model includes: collecting historical fault sample data and normal operation sample data, the historical fault sample data being labeled with fault type labels including overvoltage fault, overcurrent fault, high temperature fault, high humidity fault and load abnormality fault; performing time series division on the sample data to generate sample windows of a fixed length; extracting time sequence features of the sample windows by using a long short-term memory network and giving higher weights to key features through an attention mechanism; training the model based on a cross-entropy loss function until the anomaly recognition accuracy of the model on a validation set exceeds a preset threshold.

[0010] Optionally, the emergency level signal includes three levels: level one indicates that the load device is in a non-core task operation state and allows short-time interruption; level two indicates that the load device is in a core task operation state but has fault tolerance buffer capability; and level three indicates that the load device is in a critical task operation state and interruption may cause data loss or device damage.

[0011] Optionally, the decision logic of the decision tree model includes: If the real-time anomaly probability value is in (the first preset threshold, the second preset threshold) and the emergency level signal is level one, the protection level is determined as a pre-warning level; If the real-time anomaly probability value is in (the first preset threshold, the second preset threshold) and the emergency level signal is level two or level three, the protection level is determined as a current limiting level; If the real-time anomaly probability value is in (the second preset threshold, the third preset threshold) and the emergency level signal is level one or level two, the protection level is determined as a short-time power-off level; If the real-time anomaly probability value is greater than or equal to the third preset threshold or the emergency level signal is level three and the real-time anomaly probability value is greater than or equal to the first preset threshold, the protection level is determined as a complete power-off level; wherein the first preset threshold < the second preset threshold < the third preset threshold.

[0012] Optionally, the protection operation performed by the protection execution module includes: The pre-warning level: the power supply sends an audible and visual alarm signal and maintains normal power supply; The current limiting level: the power supply output current is dynamically adjusted to a safe threshold range, and the power of a non-core power supply loop is reduced; The short-time power-off level: the power supply is cut off for 0.5-3 seconds and then restored, triggering the self-protection mechanism of the load device; The complete power-off level: the main power supply loop is immediately cut off, and the standby power supply supplies power to the key storage components of the load device until the data saving is completed.

[0013] Optionally, the method further includes: Within a preset recovery period after the protection operation is performed, the recovery data is continuously collected by the multi-dimensional sensor group, if the recovery data meets the normal operation condition for N consecutive times (N is an integer greater than or equal to 5), the protection execution module is controlled to release the protection state and restore normal power supply; otherwise, the protection state is maintained and the prompt information level is upgraded.

[0014] Optionally, the communication module supports 5G and wired Ethernet dual-mode communication, and when the protection operation is performed, the prompt information is preferentially sent through the wired Ethernet; if the wired communication is interrupted, the communication mode is automatically switched to 5G.

[0015] Optionally, the preset weight distribution rule adopts a dynamic adjustment mechanism, specifically including the following: first, based on the type of the load device, a basic weight matrix is established, wherein for precision electronic devices (such as servers, medical instruments), the basic weights of voltage data and current data are set to 0.3 and 0.25 respectively, the weight of temperature data is 0.2, the weight of humidity data is 0.1, and the weight of load state data is 0.15; for power equipment (such as motors, pumps), the weight of current data is increased to 0.35, the weight of temperature data is 0.25, the weight of voltage data is 0.2, the weight of humidity data is 0.05, and the weight of load state data is 0.15; for ordinary office equipment, the weight of voltage data is 0.2, the weight of current data is 0.2, the weight of temperature data is 0.15, the weight of humidity data is 0.1, and the weight of load state data is 0.35. Second, a real-time feedback adjustment factor is introduced, when the running state data fluctuation amplitude of the load device in 5 consecutive sampling periods exceeds ±10%, the weight of the load state data is automatically increased by 0.05 (not more than 0.4), and the weights of other dimension data are reduced in proportion; when the internal temperature of the power supply exceeds 80% of the safety threshold for 3 consecutive sampling periods, the weight of the temperature data is increased by 0.08 (not more than 0.35), and the weights of humidity and voltage data are reduced synchronously. In addition, a power supply priority correction coefficient is also set, for A-level priority load (such as emergency lighting, core database), the weights of all safety-related data (voltage, current, temperature) are multiplied by a coefficient of 1.2; for C-level priority load (such as ordinary lighting, printer), the weights of safety-related data are multiplied by a coefficient of 0.8. Finally, the adjusted weights of each dimension are mapped to an interval with a total sum of 1 through a normalization formula, ensuring the calculation stability of the standardized data matrix. The dynamic weight distribution mechanism can adaptively adjust the data priority according to the load characteristics and real-time running state, solving the problem of insufficient adaptability of fixed weight in complex scenarios and improving the accuracy of anomaly detection.

[0016] Optionally, an incremental learning mechanism can be introduced into the anomaly detection model during the training process, which specifically includes the following steps: first, build a basic training set containing more than 100,000 historical fault samples and normal operation samples, wherein the fault samples cover 12 common fault types such as overpressure, overcurrent, high temperature, excessive humidity, and load mutation, and the sample size of each fault type is not less than 5000, and the ratio of normal samples to fault samples is controlled at 3:1. Second, after the initial training of the model is completed, set an online learning window, and trigger an incremental training every 500 new running data (including normal running data and new fault data). Before incremental training, clean the new data and remove noise data (such as sensor instantaneous false alarm and abnormal value caused by communication interference), and align the data in each dimension through time stamp to ensure time consistency. Then, use the knowledge distillation technology to retain the historical knowledge learned by the model: use the original model as the teacher model and the newly trained sub-model as the student model, minimize the output difference between the student model and the teacher model on the historical samples, and avoid catastrophic forgetting. At the same time, set a higher training weight (1.5 times of the ordinary sample) for the new fault type samples to accelerate the model's learning of new fault patterns. In addition, a model verification mechanism is set, after each incremental training, an independent verification set (containing 30% of historical samples and 70% of new samples) is used to evaluate the model performance, if the anomaly recognition accuracy decreases by more than 5% compared with the previous version, the incremental update is abandoned and the model is rolled back to the previous version. Finally, periodically (every 3 months) retrain the model, integrate all historical data and new data, and optimize the model parameters. The incremental learning mechanism enables the model to continuously adapt to new operating scenarios and fault patterns, solving the performance degradation problem of traditional static models in long-term use, and prolonging the effective life cycle of the model.

[0017] Optionally, the decision tree model adopts a multi-factor hierarchical decision logic, which specifically includes the following hierarchical structure: the first layer is the cross-judgment of abnormal probability and emergency level, which divides the real-time abnormal probability value into four intervals: [first preset threshold, 0.4), [0.4, 0.6), [0.6, 0.8), [0.8, 1.0], and the three-level division of the emergency level signal is refined into five levels: level 1 (non-core task, interruptable for more than 30 minutes), level 2 (non-core task, only interruptable for 5-30 minutes), level 3 (core task, tolerable for 1-5 minutes of interruption), level 4 (core task, only tolerable for 1 minute of interruption), and level 5 (critical task, zero interruption tolerance). The second layer introduces a time decay factor. If the abnormal probability value rises by more than 0.3 within 10 minutes, the protection level is automatically increased by one level; if the abnormal probability value remains stable or decreases within 5 minutes, the protection level is reduced by one level (the lowest is the warning level). The third layer combines the historical failure records of the load device. If the same type of abnormality has occurred to the load in the past 30 days and caused damage to the device, the current protection level is increased by one level; if the same type of abnormality has been recovered after warning in the past 30 days, the current protection level is reduced by one level. The fourth layer considers the power redundancy capability. If the intelligent power supply is connected to a standby power supply circuit and the standby power supply capacity meets more than 80% of the power demand of the load, the protection level can be appropriately reduced (such as adjusting the complete power-off level to the short-time power-off level) when determining the protection level; if the standby power supply fails or the capacity is insufficient, the protection level is maintained or increased. The final decision logic is as follows: when the abnormal probability is in [first preset threshold, 0.4) and the emergency level is 1-2, it is determined as the warning level; when the abnormal probability is in [first preset threshold, 0.4) and the emergency level is 3-5, or the abnormal probability is in [0.4, 0.6) and the emergency level is 1-2, it is determined as the current limiting level; when the abnormal probability is in [0.4, 0.6) and the emergency level is 3-4, or the abnormal probability is in [0.6, 0.8) and the emergency level is 1-3, it is determined as the short-time power-off level; when the abnormal probability is in [0.6, 0.8) and the emergency level is 4-5, or the abnormal probability is in [0.8, 1.0] and the emergency level is 1-5, it is determined as the complete power-off level. This multi-factor hierarchical decision logic considers real-time state, change trend, historical record and system redundancy, avoids the one-sidedness of single-factor decision, and makes the determination of protection level more suitable for actual scene requirements.

[0018] Optionally, the execution protection module adopts a hierarchical linkage mechanism when performing the protection operation, which specifically includes the following linkage logic: for early warning level protection, the execution module first controls the indicator light built in the power supply to switch to a yellow flashing state, the buzzer emits a prompt sound at a frequency of 1 Hz, and at the same time, pushes a text early warning information to the associated terminal through the communication module, including the abnormal dimension (such as "temperature is too high" "current fluctuation"), the current value and the safety threshold range; secondly, start the internal cooling fan (if the temperature data is abnormal) or the dehumidification device (if the humidity data is abnormal), actively adjust the power supply operating environment; finally, record the early warning occurrence time, duration and related data curve, and form an early warning log archive. For current limiting level protection, on the basis of the early warning level operation, the execution module dynamically adjusts the output current through the PWM (pulse width modulation) technology, limits the current in the range of 80%-90% of the rated current of the load, and at the same time, cuts off the non-core power supply branch of the load device (such as USB interface, expansion module), to preferentially guarantee the power supply of core components; if the current limiting lasts for 5 minutes and the abnormal probability does not decrease, automatically trigger the load device's frequency reduction instruction (sent through the communication module), to reduce its power consumption. For short-time power-off level protection, the execution module first sends a pre-notification of "power off in 3 seconds" to the load device through the communication module, so that it enters a data caching state; during power-off, the internal capacitor energy storage unit of the power supply is started to provide maintenance power supply for the real-time clock, cache chip and other key components of the load; after the power supply is restored, a restart instruction is sent, and the startup state of the load device is monitored; if 3 times of restart fail, the execution module automatically upgrades to complete power-off level. For complete power-off level protection, the execution module immediately cuts off the relay of the main power supply circuit, and at the same time, triggers the standby power supply switching, the standby power supply only supplies power to the storage module (such as hard disk, memory) of the load device, and the duration does not exceed 30 seconds (to ensure that the data saving is completed); then, all unnecessary power supply circuits are turned off, and only the operating power supply of the controller and the sensor is reserved; finally, an emergency alarm information containing a fault code is sent to the associated terminal, and the communication module is linked with the industry or operation and maintenance system to trigger an on-site maintenance work order. The hierarchical linkage mechanism ensures the hierarchy and continuity of the protection operation, while ensuring the safety of the device, and maximizes the impact on the load operation, especially suitable for scenes with high continuity requirements.

[0019] The second aspect provides an intelligent power supply for supplying power to a load device; the intelligent power supply is configured to perform the method of any one of claims 1-8.

[0020] The third aspect provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method of the first aspect.

[0021] The fourth aspect provides a computer program product, comprising a non-transitory computer-readable storage medium storing a computer program, the computer program being executed by a processor to implement the method according to the first aspect.

[0022] The multi-dimensional sensor group (covering voltage, current, temperature, humidity and load state) is adopted to collect real-time data in the embodiment of the application, which breaks through the limitation of traditional power protection relying on only single parameter of voltage and current. By integrating the running environment (temperature and humidity) of the power supply itself and the real-time state (such as running task and core function) of the load device, a more complete "power supply-load" system running image is constructed. For example, in a high temperature and high humidity environment, even if the voltage and current are not out of limit, the risk of insulation aging can be predicted through the environmental parameters; combined with the CPU occupancy rate, task priority and other data of the load, "normal power fluctuation" and "abnormal overload" can be distinguished, and misjudgment is reduced. At the same time, based on the anomaly detection model of long short-term memory network (LSTM) and attention mechanism, the time sequence correlation characteristics (such as the slow rising trend of temperature before the sudden increase of current) of multi-dimensional data can be captured, and higher weight can be given to key characteristics (such as voltage stability of precision equipment), so that the accuracy of anomaly recognition is improved by more than 30% compared with the traditional threshold method, and the early warning ability for gradual faults (such as voltage drift caused by component aging) is significantly enhanced.

[0023] The dynamic adaptation and fine method of protection strategy constructs a "abnormal probability-load emergency degree" two-dimensional protection level division mechanism by introducing the emergency degree signal of the load device and the decision tree model, which replaces the traditional "one-size-fits-all" power-off protection mode. For example, when the same abnormal probability is detected, the device running non-core tasks (such as ordinary office software) only triggers a warning, while the device executing critical tasks (such as real-time monitoring of medical equipment) immediately starts current limiting or short-time power-off, which not only avoids unnecessary downtime loss, but also ensures the safety of high-priority loads. In addition, the protection level is refined into four levels of warning, current limiting, short-time power-off and complete power-off, which can respond in a step-by-step manner according to the severity of the abnormality: when the abnormality is slight, the warning prompts manual intervention, when the abnormality is moderate, the output parameters are dynamically adjusted, and when the abnormality is serious, the power supply is cut off, so that the protection operation is more suitable for the actual risk level and reduces the production or service interruption caused by excessive protection.

[0024] The introduction of a communication module enhances the system's collaboration and traceability, enabling bidirectional interaction between the intelligent power supply, load devices, and associated terminals. On one hand, the power supply can obtain real-time status feedback from the load (such as urgency level), allowing protection decisions to better align with load requirements. On the other hand, execution information of protection operations (anomaly type, measures, and time) can be pushed to the terminal in real time, facilitating remote monitoring and intervention by maintenance personnel. For example, data center maintenance personnel can receive overcurrent warnings from a server in real time via the terminal and, combined with the "data backup in progress" signal from the load feedback, remotely adjust the protection level to current limiting rather than power outage to prevent data loss. Simultaneously, complete operation logs (anomaly data, decision basis, and execution results) provide data support for post-incident traceability, facilitating analysis of fault causes, optimization of protection strategies, and forming a closed-loop management system of "detection-decision-execution-review."

[0025] The method for expanding the adaptability and flexibility of protection scenarios, through multi-dimensional data acquisition and dynamic weight allocation (implicit in the weight rules of normalization processing), can adapt to the different characteristics of various types of loads (such as precision electronic equipment, power equipment, and office equipment). For example, for motor loads, the focus is on current and temperature parameters to prevent overload burnout; for server loads, priority is given to monitoring voltage stability and load status to ensure data security. This adaptability eliminates the need for smart power supplies to redesign protection logic for specific loads, significantly improving product versatility. Furthermore, the hierarchical operation of protection modules (such as capacitor energy storage support during short-term power outages and backup power switching during complete power outages) can meet the continuity requirements in different scenarios, making it particularly suitable for fields sensitive to power interruptions, such as medical and industrial control. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram illustrating an application scenario of a power intelligent protection and control method provided in an embodiment of this application. Figure 2 This is a flowchart illustrating a power intelligent protection and control method provided in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] The terms "first," "second," "third," and "fourth," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to these processes, methods, products, or devices.

[0030] In this document, the term "embodiment" means that a particular feature, result, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] See Figure 1 , Figure 1 This is a schematic diagram illustrating an application scenario of the intelligent power protection and control method provided in the embodiments of this application. For example... Figure 1 As shown, the intelligent power supply is connected to the load device via wires and is mainly used to supply power to the load device.

[0032] See Figure 2 , Figure 2 This is a flowchart illustrating a power supply intelligent protection and control method provided in an embodiment of this application. The method is applied to the controller of an intelligent power supply used to supply power to a load device. The intelligent power supply further includes a multi-dimensional sensor group, a communication module, and an execution protection module. The multi-dimensional sensor group includes a temperature sensor, a humidity sensor, a voltage sensor, a current sensor, and a load status sensor. The method comprises: S201. Collect multi-dimensional real-time data within a preset period through the multi-dimensional sensor group. The multi-dimensional real-time data includes real-time voltage data, real-time current data, internal ambient temperature data of the power supply, external ambient humidity data of the power supply, and real-time operating status data of the load device. S202. The multidimensional real-time data is normalized based on a preset weight allocation rule to obtain a standardized data matrix. The weight allocation rule is set according to the type of load device and power supply priority. S203. Input the standardized data matrix into the pre-trained anomaly detection model and output the real-time anomaly probability value. The anomaly detection model is a deep learning model based on the fusion of long short-term memory network and attention mechanism, and is trained by historical fault sample data and normal operation sample data. S204. If the real-time anomaly probability value is greater than the first preset threshold, the urgency signal fed back by the load device is obtained through the communication module. The urgency signal is generated by the load device according to its own core function operation status. S205. Based on the real-time anomaly probability value and the urgency signal, determine the protection level through a preset decision tree model. The protection level includes early warning level, current limiting level, short-term power outage level and complete power outage level. S206. Control the execution protection module to perform protection operations corresponding to the protection level, and send a prompt message containing the exception type and protection measures to the associated terminal through the communication module.

[0033] Optionally, the real-time operating status data of the load device includes the CPU utilization rate, memory usage rate, core component temperature, and priority identifier of the currently running task.

[0034] Optionally, the normalization process includes: mapping real-time data of each dimension to the [0,1] interval, wherein the mapping reference for voltage data is the rated voltage range of the load device, the mapping reference for current data is the rated current range of the load device, the mapping reference for temperature data is the safe operating temperature range of the power supply, and the mapping reference for humidity data is the safe operating humidity range of the power supply.

[0035] Optionally, the training process of the anomaly detection model includes: Historical fault sample data and normal operation sample data are collected. The historical fault sample data is labeled with fault type tags, including overvoltage fault, overcurrent fault, high temperature fault, excessive humidity fault, and abnormal load fault. The sample data is divided into time series to generate a fixed-length sample window. The temporal features of the sample window are extracted using a long short-term memory network, and key features are given higher weights through an attention mechanism. The model is trained based on the cross-entropy loss function until the model's anomaly recognition accuracy on the validation set exceeds a preset threshold.

[0036] Optionally, the urgency level signal includes three levels: Level 1 indicates that the load device is in a non-core task operation state, which allows for short-term interruption; Level 2 indicates that the load device is in a core task operation state, but has fault-tolerant buffering capabilities; Level 3 indicates that the load device is in a critical task operation state, and interruption may lead to data loss or device damage.

[0037] Optionally, the decision logic of the decision tree model includes: If the real-time anomaly probability value is within (the first preset threshold and the second preset threshold) and the urgency level signal is level one, the protection level is determined to be the warning level. If the real-time anomaly probability value is within (the first preset threshold and the second preset threshold) and the urgency signal is level two or three, the protection level is determined to be the current limiting level. If the real-time anomaly probability value is within [the second preset threshold, the third preset threshold) and the emergency level signal is level one or level two, the protection level is determined to be short-time power outage level. If the real-time anomaly probability value is greater than or equal to the third preset threshold, or if the emergency level signal is level three and the real-time anomaly probability value is greater than or equal to the first preset threshold, the protection level is determined to be the complete power outage level; wherein, the first preset threshold is less than the second preset threshold and the third preset threshold.

[0038] Optionally, the protection operations performed by the protection module include: Early warning level: The control power supply issues an audible and visual alarm signal to maintain normal power supply; Current limiting level: Dynamically adjusts the power supply output current to a safe threshold range, while reducing the power of non-core power supply circuits; Short-term power outage level: Power supply is restored after the power output is cut off for 0.5-3 seconds, triggering the self-protection mechanism of the load equipment; Complete power failure level: Immediately disconnect the main power supply circuit and start the backup power supply to power the critical storage components of the load device until the data is saved.

[0039] Optionally, the method further includes: Within the preset recovery period after the protection operation is performed, recovery data is continuously collected by a multi-dimensional sensor group. If the recovery data meets the normal operation conditions N times consecutively (N is an integer ≥ 5), the protection module is controlled to release the protection state and restore normal power supply; otherwise, the protection state is maintained and the prompt information level is upgraded.

[0040] Optionally, the communication module supports dual-mode communication of 5G and wired Ethernet. When performing protection operations, it prioritizes sending prompt information via wired Ethernet; if wired communication is interrupted, it automatically switches to 5G communication mode.

[0041] Optionally, the preset weight allocation rule adopts a dynamic adjustment mechanism, specifically including the following: First, a basic weight matrix is ​​established based on the type of load device. For precision electronic equipment (such as servers and medical instruments), the basic weights of voltage data and current data are set to 0.3 and 0.25, respectively, the weight of temperature data is 0.2, the weight of humidity data is 0.1, and the weight of load status data is 0.15. For power equipment (such as motors and pumps), the weight of current data is increased to 0.35, the weight of temperature data is 0.25, the weight of voltage data is 0.2, the weight of humidity data is 0.05, and the weight of load status data is 0.15. For ordinary office equipment, the weights of voltage data, current data, temperature data, humidity data, and load status data are 0.35. Secondly, a real-time feedback adjustment factor is introduced. When the fluctuation range of the operating status data of the load device exceeds ±10% within 5 consecutive sampling cycles, the weight of the load status data is automatically increased by 0.05 (maximum not exceeding 0.4), while the weight of other dimensions of data is reduced proportionally. When the internal temperature of the power supply exceeds 80% of the safety threshold for 3 consecutive sampling cycles, the weight of the temperature data is increased by 0.08 (maximum not exceeding 0.35), while the weight of humidity and voltage data is reduced simultaneously. In addition, a power supply priority correction coefficient is set. For Class A priority loads (such as emergency lighting and core databases), the weight of all safety-related data (voltage, current, temperature) is multiplied by a coefficient of 1.2; for Class C priority loads (such as general lighting and printers), the weight of safety-related data is multiplied by a coefficient of 0.8. Finally, the adjusted weights of each dimension are mapped to a range where the sum is 1 through a normalization formula to ensure the calculation stability of the standardized data matrix. This dynamic weight allocation mechanism can adaptively adjust data priorities according to load characteristics and real-time operating status, solving the problem of insufficient adaptability of fixed weights in complex scenarios and improving the accuracy of anomaly detection.

[0042] Optionally, an incremental learning mechanism can be introduced during the training of the anomaly detection model, specifically including the following steps: First, construct a basic training set containing more than 100,000 historical fault samples and normal operation samples. The fault samples cover 12 common fault types, including overvoltage, overcurrent, high temperature, excessive humidity, and sudden load changes. Each fault type has at least 5,000 samples, and the ratio of normal samples to fault samples is controlled at 3:1. Second, after the initial model training is completed, an online learning window is set up, triggering incremental training every 500 new operational data points (including normal operation data and newly added fault data). Before incremental training, the new data is cleaned to remove noisy data (such as instantaneous false alarms from sensors or outliers caused by communication interference), and the data in each dimension is aligned using timestamps to ensure temporal consistency. Then, knowledge distillation technology is used to retain the historical knowledge learned by the model: the original model is used as the teacher model, and the newly trained sub-model is used as the student model. By minimizing the output difference between the student model and the teacher model on historical samples, catastrophic forgetting is avoided. Simultaneously, higher training weights (1.5 times that of ordinary samples) are assigned to newly added fault type samples to accelerate the model's learning of new fault modes. Furthermore, a model validation mechanism is implemented: after each incremental training iteration, a separate validation set (containing 30% historical samples and 70% new samples) is used to evaluate model performance. If the anomaly identification accuracy drops by more than 5% compared to the previous version, the incremental update is abandoned, and the model reverts to the previous version. Finally, the model is periodically (every 3 months) fully retrained, integrating all historical and new data to optimize model parameters. This incremental learning mechanism enables the model to continuously adapt to new operating scenarios and fault modes, solving the performance degradation problem of traditional static models over long-term use and extending the model's effective lifespan.

[0043] Optionally, the decision tree model adopts a multi-factor hierarchical decision-making logic, specifically including the following hierarchical structure: The first layer is a cross-judgment of anomaly probability and urgency level, dividing the real-time anomaly probability value into four intervals: [first preset threshold, 0.4), [0.4, 0.6), [0.6, 0.8), [0.8, 1.0]. The three-level division of urgency level signal is refined into five levels: Level 1 (non-core task, can be interrupted for more than 30 minutes), Level 2 (non-core task, can only be interrupted for 5-30 minutes), Level 3 (core task, can tolerate 1-5 minutes of interruption), Level 4 (core task, can only tolerate interruption within 1 minute), and Level 5 (critical task, zero interruption tolerance). The second layer introduces a time decay factor. If the anomaly probability value increases by more than 0.3 within 10 minutes, the protection level is automatically increased by one level; if the anomaly probability value remains stable or decreases within 5 minutes, the protection level is decreased by one level (the lowest being the warning level). The third layer considers the historical fault records of the load equipment. If the load has experienced a similar anomaly within the past 30 days that resulted in equipment damage, the current protection level is increased by one level; if a similar anomaly recovered on its own after being alerted within the past 30 days, the current protection level is decreased by one level. The fourth layer considers power redundancy. If the intelligent power supply is connected to a backup power supply circuit and the backup power supply capacity meets more than 80% of the load's power demand, the protection level can be appropriately reduced when determining the protection level (e.g., adjusting the complete power outage level to the short-term power outage level); if the backup power supply fails or its capacity is insufficient, the protection level is maintained or increased. The final decision-making logic is as follows: When the anomaly probability is within [first preset threshold, 0.4) and the urgency level is 1-2, it is determined to be at the warning level; when the anomaly probability is within [first preset threshold, 0.4) and the urgency level is 3-5, or when the anomaly probability is within [0.4, 0.6) and the urgency level is 1-2, it is determined to be at the current limiting level; when the anomaly probability is within [0.4, 0.6) and the urgency level is 3-4, or when the anomaly probability is within [0.6, 0.8) and the urgency level is 1-3, it is determined to be at the short-term power outage level; when the anomaly probability is within [0.6, 0.8) and the urgency level is 4-5, or when the anomaly probability is within [0.8, 1.0] and the urgency level is 1-5, it is determined to be at the complete power outage level. This multi-factor hierarchical decision-making logic comprehensively considers real-time status, changing trends, historical records, and system redundancy, avoiding the one-sidedness of single-factor decision-making, and making the determination of the protection level more in line with the actual scenario requirements.

[0044] Optionally, the protection module employs a tiered linkage mechanism when performing protection operations, specifically including the following linkage logic: For warning-level protection, the execution module first controls the power supply's built-in indicator light to flash yellow, and the buzzer emits a 1Hz warning sound. Simultaneously, it pushes text warning information to associated terminals via the communication module, including the abnormal dimension (such as "high temperature" or "current fluctuation"), the current value, and the safety threshold range. Secondly, it starts the internal cooling fan (if the temperature data is abnormal) or the dehumidifier (if the humidity data is abnormal) to actively adjust the power supply's operating environment. Finally, it records the warning occurrence time, duration, and related data curves to form a warning log for archiving. For current-limiting protection, based on the warning-level operation, the execution module dynamically adjusts the output current using PWM (Pulse Width Modulation) technology, limiting the current to 80%-90% of the load's rated current. At the same time, it cuts off non-core power supply branches of the load device (such as USB interfaces and expansion modules) to prioritize power supply to core components. If the abnormal probability does not decrease after the current limiting continues for 5 minutes, it automatically triggers a frequency reduction command for the load device (sent via the communication module) to reduce its power consumption. For short-term power outage protection, the execution module first sends a pre-notification of "power outage in 3 seconds" to the load device via the communication module, causing it to enter a data buffer state. During the power outage, the internal capacitor energy storage unit of the power supply is activated to provide sustained power to critical components such as the load's real-time clock and cache chip. After power is restored, a restart command is sent, and the startup status of the load device is monitored. If three restarts fail, the system automatically escalates to a complete power outage. For complete power outage protection, the execution module immediately disconnects the relays in the main power supply circuit and triggers a backup power switch. The backup power supply only powers the load device's storage modules (such as hard drives and memory) for no more than 30 seconds (ensuring data saving is complete). Subsequently, all unnecessary power supply circuits are shut down, leaving only the controller and sensors powered. Finally, an emergency alarm message containing a fault code is sent to the associated terminal, and the communication module is used to link with the property management or maintenance system to trigger an on-site repair work order. This hierarchical linkage mechanism ensures the hierarchy and continuity of protection operations, minimizing the impact on load operation while ensuring equipment safety, making it particularly suitable for scenarios with high continuity requirements.

[0045] The second aspect provides an intelligent power supply for supplying power to a load device; the intelligent power supply is used to perform some or all of the steps of any one of the methods in the embodiments of this application.

[0046] For example, a smart power supply is used to collect multi-dimensional real-time data within a preset period through the multi-dimensional sensor group. The multi-dimensional real-time data includes real-time voltage data, real-time current data, internal ambient temperature data, external ambient humidity data, and real-time operating status data of the load device at the power supply output terminal. The multidimensional real-time data is normalized based on a preset weight allocation rule to obtain a standardized data matrix. The weight allocation rule is set according to the type of load device and power supply priority. The standardized data matrix is ​​input into a pre-trained anomaly detection model, which outputs real-time anomaly probability values. The anomaly detection model is a deep learning model based on the fusion of long short-term memory network and attention mechanism, and is trained using historical fault sample data and normal operation sample data. If the real-time anomaly probability value is greater than the first preset threshold, the urgency signal fed back by the load device is obtained through the communication module. The urgency signal is generated by the load device based on its own core function operation status. Based on the real-time anomaly probability value and the urgency signal, the protection level is determined through a preset decision tree model. The protection level includes early warning level, current limiting level, short-term power outage level and complete power outage level. The system controls the execution protection module to perform protection operations corresponding to the protection level, and sends a prompt message containing the anomaly type and protection measures to the associated terminal through the communication module.

[0047] Optionally, the real-time operating status data of the load device includes the CPU utilization rate, memory usage rate, core component temperature, and priority identifier of the currently running task.

[0048] Optionally, the normalization process includes: mapping real-time data of each dimension to the [0,1] interval, wherein the mapping reference for voltage data is the rated voltage range of the load device, the mapping reference for current data is the rated current range of the load device, the mapping reference for temperature data is the safe operating temperature range of the power supply, and the mapping reference for humidity data is the safe operating humidity range of the power supply.

[0049] Optionally, the training process of the anomaly detection model includes: Historical fault sample data and normal operation sample data are collected. The historical fault sample data is labeled with fault type tags, including overvoltage fault, overcurrent fault, high temperature fault, excessive humidity fault, and abnormal load fault. The sample data is divided into time series to generate a fixed-length sample window. The temporal features of the sample window are extracted using a long short-term memory network, and key features are given higher weights through an attention mechanism. The model is trained based on the cross-entropy loss function until the model's anomaly recognition accuracy on the validation set exceeds a preset threshold.

[0050] Optionally, the urgency level signal includes three levels: Level 1 indicates that the load device is in a non-core task operation state, which allows for short-term interruption; Level 2 indicates that the load device is in a core task operation state, but has fault-tolerant buffering capabilities; Level 3 indicates that the load device is in a critical task operation state, and interruption may lead to data loss or device damage.

[0051] Optionally, the decision logic of the decision tree model includes: If the real-time anomaly probability value is within (the first preset threshold and the second preset threshold) and the urgency level signal is level one, the protection level is determined to be the warning level. If the real-time anomaly probability value is within (the first preset threshold and the second preset threshold) and the urgency signal is level two or three, the protection level is determined to be the current limiting level. If the real-time anomaly probability value is within [the second preset threshold, the third preset threshold) and the emergency level signal is level one or level two, the protection level is determined to be short-time power outage level. If the real-time anomaly probability value is greater than or equal to the third preset threshold, or if the emergency level signal is level three and the real-time anomaly probability value is greater than or equal to the first preset threshold, the protection level is determined to be the complete power outage level; wherein, the first preset threshold is less than the second preset threshold and the third preset threshold.

[0052] Optionally, the protection operations performed by the protection module include: Early warning level: The control power supply issues an audible and visual alarm signal to maintain normal power supply; Current limiting level: Dynamically adjusts the power supply output current to a safe threshold range, while reducing the power of non-core power supply circuits; Short-term power outage level: Power supply is restored after the power output is cut off for 0.5-3 seconds, triggering the self-protection mechanism of the load equipment; Complete power failure level: Immediately disconnect the main power supply circuit and start the backup power supply to power the critical storage components of the load device until the data is saved.

[0053] Optionally, the method further includes: Within the preset recovery period after the protection operation is performed, recovery data is continuously collected by a multi-dimensional sensor group. If the recovery data meets the normal operation conditions N times consecutively (N is an integer ≥ 5), the protection module is controlled to release the protection state and restore normal power supply; otherwise, the protection state is maintained and the prompt information level is upgraded.

[0054] Optionally, the communication module supports dual-mode communication of 5G and wired Ethernet. When performing protection operations, it prioritizes sending prompt information via wired Ethernet; if wired communication is interrupted, it automatically switches to 5G communication mode.

[0055] Optionally, the preset weight allocation rule adopts a dynamic adjustment mechanism, specifically including the following: First, a basic weight matrix is ​​established based on the type of load device. For precision electronic equipment (such as servers and medical instruments), the basic weights of voltage data and current data are set to 0.3 and 0.25, respectively, the weight of temperature data is 0.2, the weight of humidity data is 0.1, and the weight of load status data is 0.15. For power equipment (such as motors and pumps), the weight of current data is increased to 0.35, the weight of temperature data is 0.25, the weight of voltage data is 0.2, the weight of humidity data is 0.05, and the weight of load status data is 0.15. For ordinary office equipment, the weights of voltage data, current data, temperature data, humidity data, and load status data are 0.35. Secondly, a real-time feedback adjustment factor is introduced. When the fluctuation range of the operating status data of the load device exceeds ±10% within 5 consecutive sampling cycles, the weight of the load status data is automatically increased by 0.05 (maximum not exceeding 0.4), while the weight of other dimensions of data is reduced proportionally. When the internal temperature of the power supply exceeds 80% of the safety threshold for 3 consecutive sampling cycles, the weight of the temperature data is increased by 0.08 (maximum not exceeding 0.35), while the weight of humidity and voltage data is reduced simultaneously. In addition, a power supply priority correction coefficient is set. For Class A priority loads (such as emergency lighting and core databases), the weight of all safety-related data (voltage, current, temperature) is multiplied by a coefficient of 1.2; for Class C priority loads (such as general lighting and printers), the weight of safety-related data is multiplied by a coefficient of 0.8. Finally, the adjusted weights of each dimension are mapped to a range where the sum is 1 through a normalization formula to ensure the calculation stability of the standardized data matrix. This dynamic weight allocation mechanism can adaptively adjust data priorities according to load characteristics and real-time operating status, solving the problem of insufficient adaptability of fixed weights in complex scenarios and improving the accuracy of anomaly detection.

[0056] Optionally, an incremental learning mechanism can be introduced during the training of the anomaly detection model, specifically including the following steps: First, construct a basic training set containing more than 100,000 historical fault samples and normal operation samples. The fault samples cover 12 common fault types, including overvoltage, overcurrent, high temperature, excessive humidity, and sudden load changes. Each fault type has at least 5,000 samples, and the ratio of normal samples to fault samples is controlled at 3:1. Second, after the initial model training is completed, an online learning window is set up, triggering incremental training every 500 new operational data points (including normal operation data and newly added fault data). Before incremental training, the new data is cleaned to remove noisy data (such as instantaneous false alarms from sensors or outliers caused by communication interference), and the data in each dimension is aligned using timestamps to ensure temporal consistency. Then, knowledge distillation technology is used to retain the historical knowledge learned by the model: the original model is used as the teacher model, and the newly trained sub-model is used as the student model. By minimizing the output difference between the student model and the teacher model on historical samples, catastrophic forgetting is avoided. Simultaneously, higher training weights (1.5 times that of ordinary samples) are assigned to newly added fault type samples to accelerate the model's learning of new fault modes. Furthermore, a model validation mechanism is implemented: after each incremental training iteration, a separate validation set (containing 30% historical samples and 70% new samples) is used to evaluate model performance. If the anomaly identification accuracy drops by more than 5% compared to the previous version, the incremental update is abandoned, and the model reverts to the previous version. Finally, the model is periodically (every 3 months) fully retrained, integrating all historical and new data to optimize model parameters. This incremental learning mechanism enables the model to continuously adapt to new operating scenarios and fault modes, solving the performance degradation problem of traditional static models over long-term use and extending the model's effective lifespan.

[0057] Optionally, the decision tree model adopts a multi-factor hierarchical decision-making logic, specifically including the following hierarchical structure: The first layer is a cross-judgment of anomaly probability and urgency level, dividing the real-time anomaly probability value into four intervals: [first preset threshold, 0.4), [0.4, 0.6), [0.6, 0.8), [0.8, 1.0]. The three-level division of urgency level signal is refined into five levels: Level 1 (non-core task, can be interrupted for more than 30 minutes), Level 2 (non-core task, can only be interrupted for 5-30 minutes), Level 3 (core task, can tolerate 1-5 minutes of interruption), Level 4 (core task, can only tolerate interruption within 1 minute), and Level 5 (critical task, zero interruption tolerance). The second layer introduces a time decay factor. If the anomaly probability value increases by more than 0.3 within 10 minutes, the protection level is automatically increased by one level; if the anomaly probability value remains stable or decreases within 5 minutes, the protection level is decreased by one level (the lowest being the warning level). The third layer considers the historical fault records of the load equipment. If the load has experienced a similar anomaly within the past 30 days that resulted in equipment damage, the current protection level is increased by one level; if a similar anomaly recovered on its own after being alerted within the past 30 days, the current protection level is decreased by one level. The fourth layer considers power redundancy. If the intelligent power supply is connected to a backup power supply circuit and the backup power supply capacity meets more than 80% of the load's power demand, the protection level can be appropriately reduced when determining the protection level (e.g., adjusting the complete power outage level to the short-term power outage level); if the backup power supply fails or its capacity is insufficient, the protection level is maintained or increased. The final decision-making logic is as follows: When the anomaly probability is within [first preset threshold, 0.4) and the urgency level is 1-2, it is determined to be at the warning level; when the anomaly probability is within [first preset threshold, 0.4) and the urgency level is 3-5, or when the anomaly probability is within [0.4, 0.6) and the urgency level is 1-2, it is determined to be at the current limiting level; when the anomaly probability is within [0.4, 0.6) and the urgency level is 3-4, or when the anomaly probability is within [0.6, 0.8) and the urgency level is 1-3, it is determined to be at the short-term power outage level; when the anomaly probability is within [0.6, 0.8) and the urgency level is 4-5, or when the anomaly probability is within [0.8, 1.0] and the urgency level is 1-5, it is determined to be at the complete power outage level. This multi-factor hierarchical decision-making logic comprehensively considers real-time status, changing trends, historical records, and system redundancy, avoiding the one-sidedness of single-factor decision-making, and making the determination of the protection level more in line with the actual scenario requirements.

[0058] Optionally, the protection module employs a tiered linkage mechanism when performing protection operations, specifically including the following linkage logic: For warning-level protection, the execution module first controls the power supply's built-in indicator light to flash yellow, and the buzzer emits a 1Hz warning sound. Simultaneously, it pushes text warning information to associated terminals via the communication module, including the abnormal dimension (such as "high temperature" or "current fluctuation"), the current value, and the safety threshold range. Secondly, it starts the internal cooling fan (if the temperature data is abnormal) or the dehumidifier (if the humidity data is abnormal) to actively adjust the power supply's operating environment. Finally, it records the warning occurrence time, duration, and related data curves to form a warning log for archiving. For current-limiting protection, based on the warning-level operation, the execution module dynamically adjusts the output current using PWM (Pulse Width Modulation) technology, limiting the current to 80%-90% of the load's rated current. At the same time, it cuts off non-core power supply branches of the load device (such as USB interfaces and expansion modules) to prioritize power supply to core components. If the abnormal probability does not decrease after the current limiting continues for 5 minutes, it automatically triggers a frequency reduction command for the load device (sent via the communication module) to reduce its power consumption. For short-term power outage protection, the execution module first sends a pre-notification of "power outage in 3 seconds" to the load device via the communication module, causing it to enter a data buffer state. During the power outage, the internal capacitor energy storage unit of the power supply is activated to provide sustained power to critical components such as the load's real-time clock and cache chip. After power is restored, a restart command is sent, and the startup status of the load device is monitored. If three restarts fail, the system automatically escalates to a complete power outage. For complete power outage protection, the execution module immediately disconnects the relays in the main power supply circuit and triggers a backup power switch. The backup power supply only powers the load device's storage modules (such as hard drives and memory) for no more than 30 seconds (ensuring data saving is complete). Subsequently, all unnecessary power supply circuits are shut down, leaving only the controller and sensors powered. Finally, an emergency alarm message containing a fault code is sent to the associated terminal, and the communication module is used to link with the property management or maintenance system to trigger an on-site repair work order. This hierarchical linkage mechanism ensures the hierarchy and continuity of protection operations, minimizing the impact on load operation while ensuring equipment safety, making it particularly suitable for scenarios with high continuity requirements.

[0059] Thirdly, a computer-readable storage medium is provided that stores a computer program, which is executed by a processor to implement the method as described in the first aspect.

[0060] A fourth aspect provides a computer program product comprising a non-transitory computer-readable storage medium storing a computer program that is executed by a processor to implement the method as described in the first aspect.

[0061] The embodiments of this application employ a multi-dimensional sensor array (covering voltage, current, temperature, humidity, and load status) to collect real-time data, overcoming the limitations of traditional power supply protection that relies solely on voltage and current parameters. By integrating the power supply's own operating environment (temperature and humidity) with the real-time status of the load equipment (such as running tasks and core functions), a more complete "power supply-load" system operation profile is constructed. For example, in high-temperature and high-humidity environments, even if voltage and current do not exceed limits, the risk of insulation aging can be predicted through environmental parameters; combined with data such as the load's CPU utilization and task priority, "normal power fluctuations" and "abnormal overloads" can be distinguished, reducing false alarms. Simultaneously, the anomaly detection model based on Long Short-Term Memory (LSTM) networks and attention mechanisms can capture the temporal correlation characteristics of multi-dimensional data (such as the slow temperature rise trend before a sudden current surge) and assign higher weights to key features (such as the voltage stability of precision equipment), improving the anomaly identification accuracy by more than 30% compared to traditional threshold methods, especially significantly enhancing the early warning capability for gradual faults (such as voltage drift caused by component aging).

[0062] The method for achieving dynamic adaptation and refinement of protection strategies introduces the urgency signal of load devices and a decision tree model to construct a two-dimensional protection level classification mechanism of "anomaly probability - load urgency," replacing the traditional "one-size-fits-all" power-off protection mode. For example, when the same anomaly probability is detected, only an alert is triggered for devices running non-core tasks (such as ordinary office software), while devices performing critical tasks (such as real-time monitoring of medical equipment) are immediately initiated with current limiting or short-term power cut-off. This avoids unnecessary downtime losses and ensures the safety of high-priority loads. In addition, the protection level is refined into four levels: alert, current limiting, short-term power cut-off, and complete power cut-off. It can respond in a tiered manner according to the severity of the anomaly: manual intervention is prompted through alerts for minor anomalies, output parameters are dynamically adjusted for moderate anomalies, and power is cut off only for severe anomalies. This makes the protection operation more in line with the actual risk level and reduces production or service interruptions caused by over-protection.

[0063] The introduction of a communication module enhances the system's collaboration and traceability, enabling bidirectional interaction between the intelligent power supply, load devices, and associated terminals. On one hand, the power supply can obtain real-time status feedback from the load (such as urgency level), allowing protection decisions to better align with load requirements. On the other hand, execution information of protection operations (anomaly type, measures, and time) can be pushed to the terminal in real time, facilitating remote monitoring and intervention by maintenance personnel. For example, data center maintenance personnel can receive overcurrent warnings from a server in real time via the terminal and, combined with the "data backup in progress" signal from the load feedback, remotely adjust the protection level to current limiting rather than power outage to prevent data loss. Simultaneously, complete operation logs (anomaly data, decision basis, and execution results) provide data support for post-incident traceability, facilitating analysis of fault causes, optimization of protection strategies, and forming a closed-loop management system of "detection-decision-execution-review."

[0064] The method for expanding the adaptability and flexibility of protection scenarios, through multi-dimensional data acquisition and dynamic weight allocation (implicit in the weight rules of normalization processing), can adapt to the different characteristics of various types of loads (such as precision electronic equipment, power equipment, and office equipment). For example, for motor loads, the focus is on current and temperature parameters to prevent overload burnout; for server loads, priority is given to monitoring voltage stability and load status to ensure data security. This adaptability eliminates the need for smart power supplies to redesign protection logic for specific loads, significantly improving product versatility. Furthermore, the hierarchical operation of protection modules (such as capacitor energy storage support during short-term power outages and backup power switching during complete power outages) can meet the continuity requirements in different scenarios, making it particularly suitable for fields sensitive to power interruptions, such as medical and industrial control.

[0065] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement some or all of the steps of any of the methods described in the above method embodiments.

[0066] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments.

[0067] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0068] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0069] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0070] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0072] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0073] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0074] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A smart power supply protection method, applied to the controller of a smart power supply, the smart power supply being used to supply power to a load device, the smart power supply further comprising a multi-dimensional sensor group, a communication module, and an execution protection module, the multi-dimensional sensor group comprising a temperature sensor, a humidity sensor, a voltage sensor, a current sensor, and a load status sensor, characterized in that, The method includes: The multidimensional sensor group collects multidimensional real-time data within a preset period, including real-time voltage data, real-time current data, internal ambient temperature data, external ambient humidity data, and real-time operating status data of the load device. The multidimensional real-time data is normalized based on a preset weight allocation rule to obtain a standardized data matrix. The weight allocation rule is set according to the type of load device and power supply priority. The standardized data matrix is ​​input into a pre-trained anomaly detection model, which outputs real-time anomaly probability values. The anomaly detection model is a deep learning model based on the fusion of long short-term memory network and attention mechanism, and is trained using historical fault sample data and normal operation sample data. If the real-time anomaly probability value is greater than the first preset threshold, the urgency signal fed back by the load device is obtained through the communication module. The urgency signal is generated by the load device based on its own core function operation status. Based on the real-time anomaly probability value and the urgency signal, the protection level is determined through a preset decision tree model. The protection level includes early warning level, current limiting level, short-term power outage level and complete power outage level. The system controls the execution protection module to perform protection operations corresponding to the protection level, and sends a prompt message containing the anomaly type and protection measures to the associated terminal through the communication module.

2. The method according to claim 1, characterized in that, The real-time operating status data of the load device includes the CPU utilization rate, memory usage rate, core component temperature, and priority identifier of the currently running task.

3. The method according to claim 1, characterized in that, The normalization process includes mapping real-time data of each dimension to the [0,1] interval, wherein the mapping reference for voltage data is the rated voltage range of the load device, the mapping reference for current data is the rated current range of the load device, the mapping reference for temperature data is the safe operating temperature range of the power supply, and the mapping reference for humidity data is the safe operating humidity range of the power supply.

4. The method according to any one of claims 1 to 3, characterized in that, The training process of the anomaly detection model includes: Historical fault sample data and normal operation sample data are collected. The historical fault sample data is labeled with fault type tags, including overvoltage fault, overcurrent fault, high temperature fault, excessive humidity fault, and abnormal load fault. The sample data is divided into time series to generate a fixed-length sample window. The temporal features of the sample window are extracted using a long short-term memory network, and key features are given higher weights through an attention mechanism. The model is trained based on the cross-entropy loss function until the model's anomaly recognition accuracy on the validation set exceeds a preset threshold.

5. The method according to any one of claims 1 to 3, characterized in that, The urgency level signal includes three levels: Level 1 indicates that the load device is in a non-core task operation state, which allows for short-term interruption; Level 2 indicates that the load device is in a core task operation state, but has fault-tolerant buffering capabilities; Level 3 indicates that the load device is in a critical task operation state, and an interruption may result in data loss or equipment damage.

6. The method according to any one of claims 1 to 3, characterized in that, The decision logic of the decision tree model includes: If the real-time anomaly probability value is within (the first preset threshold and the second preset threshold) and the urgency level signal is level one, the protection level is determined to be the warning level. If the real-time anomaly probability value is within (the first preset threshold and the second preset threshold) and the urgency signal is level two or three, the protection level is determined to be the current limiting level. If the real-time anomaly probability value is within [the second preset threshold, the third preset threshold) and the emergency level signal is level one or level two, the protection level is determined to be short-time power outage level. If the real-time anomaly probability value is greater than or equal to the third preset threshold, or if the emergency level signal is level three and the real-time anomaly probability value is greater than or equal to the first preset threshold, the protection level is determined to be the complete power outage level; wherein, the first preset threshold is less than the second preset threshold and the third preset threshold.

7. The method according to any one of claims 1 to 3, characterized in that, The protection operations performed by the protection module include: Early warning level: The control power supply issues an audible and visual alarm signal to maintain normal power supply; Current limiting level: Dynamically adjusts the power supply output current to a safe threshold range, while reducing the power of non-core power supply circuits; Short-term power outage level: Power supply is restored after the power output is cut off for 0.5-3 seconds, triggering the self-protection mechanism of the load equipment; Complete power failure level: Immediately disconnect the main power supply circuit and start the backup power supply to power the critical storage components of the load device until the data is saved.

8. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Within the preset recovery period after the protection operation is performed, recovery data is continuously collected by a multi-dimensional sensor group. If the recovery data meets the normal operation conditions N times consecutively (N is an integer ≥ 5), the protection module is controlled to release the protection state and restore normal power supply; otherwise, the protection state is maintained and the prompt information level is upgraded.

9. The method according to any one of claims 1 to 8, characterized in that, The communication module supports dual-mode communication of 5G and wired Ethernet. When performing protection operations, it prioritizes sending prompt information via wired Ethernet; if wired communication is interrupted, it automatically switches to 5G communication mode.

10. A smart power supply, characterized in that, The intelligent power supply is used to power the load device; the intelligent power supply is used to perform the method as described in any one of claims 1-8.

Images