Industrial intelligent socket based on edge computing and safe energy-saving control method thereof

Abstract

This invention discloses an industrial smart socket based on edge computing and its safety and energy-saving control method. The method includes: responding to the detection of a plug being inserted into the socket or receiving a trigger signal indicating device access; while keeping the control relay open; acquiring timing data of voltage and current during the startup transient of the accessed device; extracting a multi-dimensional electrical feature vector, including at least the peak value of the startup surge current, the energy entropy of the current rising edge, and the distance measure between the current waveform calculated based on the DTW algorithm and a standard template library; inputting the vector into a local lightweight device fingerprint recognition model; outputting a device type identifier, historical operating data associated with the device type, and a corresponding confidence score; and performing a dynamic access verification process based on a preset confidence threshold. This invention can complete device identification and decision-making before the relay closes, shifting the security defense from "post-event disconnection" to "pre-event interception," and physically eliminating the risk of unauthorized device access.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to an industrial smart socket based on edge computing and its safety and energy-saving control method. Background Technology

[0002] Industrial smart sockets, serving as end-point sensing and control nodes in the Industrial Internet of Things (IIoT), are widely used in factory production lines, data centers, laboratories, unattended base stations, and other scenarios. Existing industrial smart sockets typically possess basic functions such as remote on / off control, power parameter monitoring, and overload protection. Their typical architecture involves the socket uploading collected data such as voltage, current, and power to a cloud or local server via Wi-Fi, Ethernet, or 4G modules. A host computer then analyzes the data and makes decisions before issuing control commands to the socket for execution.

[0003] However, existing technologies have at least the following shortcomings in terms of electrical safety access control: Current socket safety protection mainly relies on overload protectors or threshold judgment logic from a host computer. When a device is connected, the socket is first powered on, and the tripping is determined by continuously monitoring whether the current exceeds a preset threshold. This "power on first, monitor later, then power off" mechanism is a passive response protection. For illegally connected high-power devices, unauthorized mobile devices, or devices with potential malfunctions, protection is only triggered after power is supplied, by which time line overload, voltage drop, or even electrical fire may have already occurred. Especially in high-safety scenarios such as semiconductor cleanrooms, chemical explosion-proof areas, and military laboratories, the access of any unauthorized device is strictly prohibited, and existing technologies cannot identify and intercept the device before physical connection. Some existing solutions attempt to collect steady-state power data during device operation and upload it to the cloud for device type identification, thereby achieving electrical management. However, this approach has several drawbacks. First, cloud-based identification suffers from network transmission delays. It can take hundreds of milliseconds or even several seconds from device access to completion of identification and power-off. During this time, the device is already powered on, creating a potential safety hazard. Second, when the network is interrupted or the cloud service malfunctions, the socket is downgraded to a regular socket, completely losing its intelligent identification capabilities and rendering the security strategy ineffective.

[0004] Therefore, how to achieve millisecond-level device identification and security access decision-making at the socket before the physical connection of the device, so as to realize the pre-emptive security protection of "identification before power-on", and at the same time ensure that the security policy can still be effectively executed when the network is interrupted, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address at least one of the aforementioned technical problems, this invention provides an industrial smart socket based on edge computing and its safety and energy-saving control method.

[0006] In a first aspect, the present invention provides a safety and energy-saving control method for industrial smart sockets based on edge computing, the method comprising: Step S1: In response to detecting that a plug is inserted into a socket or receiving a trigger signal for device access, while keeping the control relay open, collect timing data of voltage and current of the accessed device during the startup transient period, wherein the startup transient period is a predetermined time window from the moment the device is accessed until the device enters steady-state operation. Step S2: Based on the edge computing unit integrated inside the socket, extract multi-dimensional electrical feature vectors from the time-series data. The multi-dimensional electrical feature vectors include at least the peak value of the initiation surge current, the energy entropy of the current rising edge, and the distance measure between the current waveform calculated based on the DTW algorithm and the standard template library. Step S3: Input the multi-dimensional electrical feature vector into the locally pre-stored lightweight device fingerprint recognition model, and output the device type identifier, historical operating data associated with the device type, and the corresponding confidence score; Step S4: Perform a dynamic admission verification process based on the confidence score and the preset confidence threshold.

[0007] Preferably, step S4, which involves performing a dynamic admission verification process based on the confidence score and a preset confidence interval, includes: Step S41: When the confidence score is greater than or equal to the high confidence threshold, the direct conduction mode is entered; the device type identifier is checked to see if it is in the whitelist of the current socket's access policy, and the operating power is assessed to see if it does not exceed the current socket's remaining capacity threshold; if the verification is successful, a conduction command is generated to control the relay to close, and the overcurrent protection threshold is dynamically configured according to the historical operating data associated with the device type. Step S42: When the confidence score is greater than or equal to the low confidence threshold and less than the high confidence threshold, enter the active detection mode: control the relay to perform a pulse switching operation of a preset duration, collect the back electromotive force characteristics of the device at the moment of pulse cut-off and the residual charge release characteristics at the second connection, update the multidimensional electrical feature vector, and return to execute step S3. Step S43: When the confidence score is less than the low confidence threshold, or when step S42 fails the verification, a rejection command is generated to keep the relay open and an alarm message is output.

[0008] Preferably, the admission strategy in step S41 further includes a thermal-electrical coupling security verification sub-step: Obtain the real-time temperature and historical temperature rise curve slope of key internal nodes of the socket; Based on the rated root mean square current corresponding to the identified equipment type and the estimated contact resistance of the current relay contact, a short-time temperature rise prediction model is constructed to obtain the predicted temperature value. The predicted temperature value is compared with the tolerance limit temperature of the insulation material. When the predicted temperature value exceeds the tolerance limit temperature of the insulation material, a rejection command is generated directly, and the socket is locked until the predicted temperature value drops below the tolerance limit temperature.

[0009] Preferably, the lightweight device fingerprint recognition model in step S3 is composed of a convolutional neural network branch and a gradient boosting decision tree branch connected in parallel, and the waveform morphology features and statistical features are extracted and then fused for decision.

[0010] Preferably, the predetermined time window mentioned in step S1 is an adaptive time window, calculated as follows: The short-time energy change rate of the current signal is calculated in real time. When the short-time energy change rate is lower than 1.5 times the background noise energy threshold for K consecutive power frequency cycles, the start-up transient is determined to have ended, and the data acquisition window is dynamically truncated as a predetermined time window.

[0011] Preferably, the method further includes: The control edge computing unit continuously records the electrical characteristic vectors of the actual operation phase during the device operation process; Calculate the Mahalanobis distance between the electrical feature vectors during actual operation and the cluster centers of the equipment types stored locally; If the Mahalanobis distance exceeds the dynamic threshold that decays over time for N consecutive times, it is determined that the device status has drifted or the model has changed; the automatic access function of the socket is suspended, marked as "pending verification", and the abnormal fragment is encrypted and uploaded to the host computer, requesting manual or cloud-based large model intervention for recalibration.

[0012] Secondly, the present invention also provides an industrial smart socket based on edge computing, comprising: The power conversion module is used to provide operating power to the internal circuitry of the socket; The plug detection module is used to detect the plug insertion status and generate a trigger signal when the plug insertion is detected; Relays and drive circuits are used to control the on / off state of sockets; Current and voltage sampling circuits are used to collect electrical characteristic data of connected devices; An edge computing unit is connected to the plug detection module, the current and voltage sampling circuit, and the relay drive circuit. The edge computing unit includes a microcontroller and an internally integrated non-volatile memory. The non-volatile memory is pre-loaded with a lightweight device fingerprint recognition model and an access strategy library. The communication module is used for data interaction with the host computer; The edge computing unit is configured to perform a method as described in the first aspect above and any possible implementation thereof.

[0013] Preferably, the industrial smart socket based on edge computing further includes a local alarm module connected to the edge computing unit; the local alarm module includes one or more of an indicator light, a buzzer, or a display screen, and is used to output local alarm information when the device fails to be recognized.

[0014] Preferably, in the edge computing-based industrial smart socket, an analog-to-digital converter is provided between the edge computing unit and the current and voltage sampling circuit, and the sampling rate of the analog-to-digital converter is not less than 10kHz.

[0015] Thirdly, the present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, the computer program including program instructions; when the program instructions are executed by the edge computing unit, the edge computing unit performs a method as described in the first aspect above and any possible implementation thereof.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) Existing technologies employ a passive protection mode of "power on first, monitor later, then power off," meaning that unauthorized devices are already powered on upon connection, posing a significant safety hazard. This invention completes device identification and decision-making before the relay closes, shifting the safety defense from "post-event disconnection" to "pre-event interception." By utilizing an identification window before the transient start-up, it completes identity verification before the device receives full power. For high-security scenarios such as semiconductor cleanrooms and chemical explosion-proof areas, this physically eliminates the risk of unauthorized device access.

[0017] 2) A confidence-based hierarchical verification mechanism is introduced, dividing the identification results into several intervals: When in the high-confidence interval, the system directly enters the continuity verification stage, quickly completing the connection of legitimate devices and minimizing identification delay. When in the low-confidence interval, an active detection mode is activated, acquiring additional feature information through pulse-based switching operations—the back electromotive force characteristic at the moment of pulse cut-off reflects the internal inductance characteristics of the device, and the residual charge release characteristic during secondary connection reflects the internal capacitance characteristics of the device. These two types of features have strong discriminative power in distinguishing between inductance-dominated and capacitance-dominated devices, and are triggered only when the confidence level is insufficient, avoiding the time overhead and contact wear caused by intrusive detection of all devices.

[0018] 3) The lightweight device fingerprint recognition model is pre-installed in the edge computing unit local to the socket. The recognition process is completed locally, unaffected by network conditions, and the recognition latency is greatly reduced. When the communication link is interrupted, the socket automatically switches to offline autonomous mode without degrading the security policy, significantly improving the robustness and reliability of security protection.

[0019] 4) By activating transient features instead of steady-state features through the acquisition device, the computational requirements for the recognition model are greatly reduced; combined with model compression technology, models such as neural networks and decision trees are lightweighted and solidified in the non-volatile memory of the socket MCU, realizing real-time inference on embedded devices with limited computing power.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the background art, the accompanying drawings used in the embodiments of the present invention or the background art will be described below.

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to illustrate the technical solutions of this disclosure.

[0023] Figure 1 A flowchart illustrating a safety and energy-saving control method for an industrial smart socket based on edge computing, provided for an embodiment of the present invention; Figure 2 This is a schematic diagram of a safety and energy-saving control system for an industrial smart socket based on edge computing, provided as an embodiment of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, 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.

[0025] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. 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.

[0026] Industrial smart sockets are an advanced version of sockets that integrate metering, communication, and control functions, essentially upgrading traditional sockets into end nodes of the Industrial Internet of Things (IIoT). Their working principle involves real-time collection of electrical parameters such as current, voltage, and power through built-in sensors, data analysis and edge computing using a microprocessor, and uploading the data to the cloud or management platform via communication modules such as Wi-Fi, 4G, or industrial bus. Simultaneously, they receive remote commands to execute on / off or policy control. Their advantages are mainly reflected in: supporting remote monitoring and automated management; accurately monitoring equipment energy consumption and operating status; possessing active safety protection mechanisms such as overload and leakage protection; and achieving refined energy scheduling through linkage with MES, EMS, and other systems, significantly improving the safety and energy efficiency of industrial production.

[0027] Existing industrial smart sockets typically possess basic functions such as remote on / off control, power parameter monitoring, and overload protection. Their typical architecture involves the socket uploading collected voltage, current, and power data to a cloud or local server via Wi-Fi, Ethernet, or 4G modules. A host computer then analyzes the data and makes decisions before issuing control commands to the socket for execution. However, the safety protection of existing sockets primarily relies on overload protectors or threshold judgment logic from the host computer. When a connected device is running, the socket first supplies power, continuously monitoring whether the current exceeds a preset threshold to determine whether to trip. This "power on first, monitor later, power off later" mechanism is a passive response protection. For illegally connected high-power devices, unauthorized mobile devices, or devices with potential malfunctions, protection is only triggered after power is supplied, by which time overload, voltage drops, or even electrical fires may have already occurred. Therefore, this invention aims to provide a safety and energy-saving control method for industrial smart sockets based on edge computing. It can realize millisecond-level device identification and security access decision-making locally on the socket before the physical connection of the device, realize the pre-emptive safety protection of "identification before power-on", and ensure that the security policy can still be effectively executed when the network is interrupted.

[0028] Please see Figure 1 , Figure 1 This is a flowchart illustrating a safety and energy-saving control method for an industrial smart socket based on edge computing, provided as an embodiment of the present invention. Figure 1 As shown, the method includes: Step S1: In response to detecting that a plug is inserted into a socket or receiving a trigger signal for device access, while keeping the control relay open, collect timing data of voltage and current of the accessed device during the startup transient period, wherein the startup transient period is a predetermined time window from the moment the device is accessed until the device enters steady-state operation. Step S2: Based on the edge computing unit integrated inside the socket, extract multi-dimensional electrical feature vectors from the time-series data. The multi-dimensional electrical feature vectors include at least the peak value of the initiation surge current, the energy entropy of the current rising edge, and the distance measure between the current waveform calculated based on the DTW algorithm and the standard template library. Step S3: Input the multi-dimensional electrical feature vector into the locally pre-stored lightweight device fingerprint recognition model, and output the device type identifier, historical operating data associated with the device type, and the corresponding confidence score; Step S4: Perform a dynamic admission verification process based on the confidence score and the preset confidence threshold.

[0029] This embodiment primarily utilizes edge computing and device fingerprint recognition to shift the security decision-making point from the cloud or host computer "after power is applied" to the local socket "before power is applied." In step S1, data acquisition is performed under physical isolation. When the device plug is inserted, the system first keeps the relay open, ensuring the circuit is physically unconducted. Then, high-precision sensors collect the extremely short initiation current and voltage waveforms generated by the device at the moment power is applied. The key to this step is "eavesdropping" on the device's electrical characteristics without power supply, providing raw data for subsequent analysis and achieving the security prerequisite of physical isolation from the source. In step S2, local features are extracted. The socket's built-in edge computing unit, such as a high-performance MCU or edge AI chip, immediately analyzes the acquired transient waveforms. It extracts multi-dimensional features such as surge peaks, energy change patterns, and waveform shape differences, transforming this continuous waveform data into a "digital fingerprint" that uniquely characterizes the device. This step is completed locally, providing millisecond-level response without waiting for network transmission. In step S3, real-time identification is achieved by inputting the feature vector extracted in the previous step into a lightweight AI model (such as a pruned neural network) pre-trained and stored in the socket's memory. This model quickly compares the current features with a known "device fingerprint database," instantly determining the type of accessing device, such as whether it's an authorized large motor or an unauthorized electric heater, and providing a confidence level for the identification result. Local model operation ensures that the identification process does not rely on external networks or servers. Finally, in step S4, the system makes a final decision based on the confidence level output by the model. If the device is identified as authorized and the confidence level is high, the relay is closed, ensuring safe power supply; if the device is identified as unknown or illegal, or the confidence level is too low, the relay remains open, and an alarm can be triggered via indicator lights or the network. The decision-making logic is executed entirely locally, achieving a final security closed loop.

[0030] Therefore, this embodiment directly overturns the traditional "power on first, monitor later" process, replacing it with a new paradigm of "sensing first, identifying later, making a decision, and finally powering on." By using the electrical characteristics of the device startup to complete identity authentication before physical connection, it can prevent unauthorized high-power devices and faulty devices from obtaining power within milliseconds, thereby fundamentally eliminating the risk of overload, short circuit, or fire that may be caused by them being powered on. At the same time, all calculations and decisions are completed locally at the socket, ensuring that the core security strategy can still be executed independently and effectively in extreme cases such as network interruption or cloud disconnection, greatly improving the reliability and security of the system.

[0031] In one embodiment, the core of step S1 is to capture the electrical characteristics of the device at the moment of connection, while ensuring that the main circuit is absolutely disconnected. Its specific implementation depends on the internal hardware architecture of the socket and the logic control of the embedded software. The main process is as follows: 1.1) Hardware Triggering and State Locking: The sockets on the receptacle panel have built-in micro-switches or photoelectric sensors. When a physical plug is inserted, the sensor immediately generates a "device access" trigger signal. Upon receiving this signal, the main control MCU first performs a critical operation: locking the drive pin of the control power supply relay, keeping it in the OFF state. This is a high-priority hardware and software dual safeguard, ensuring that any subsequent operations are performed in a power-free environment.

[0032] 1.2) Constructing the test loop and signal acquisition: The socket is internally designed with a separate, low-voltage (e.g., 12V / 24V), low-current test signal loop. This loop is physically separated from the 220V / 380V main power supply line, but through high-impedance coupling or a dedicated sensor, it can non-invasively sense the response of the connected device to a minor stimulus. Once the relay is locked in the open state, the MCU immediately closes a miniature control switch in the test loop, applying a safe, non-destructive test voltage signal to the device. A high-precision, high-speed ADC (analog-to-digital converter) sampling circuit simultaneously starts, acquiring the instantaneous voltage and current waveforms exhibited by the device in the test loop at a high frequency (e.g., 10-100kHz). This "startup transient" is typically extremely short, possibly lasting from 50 to 500 milliseconds, sufficient to cover characteristic electrical signals generated during capacitor charging, controller initialization, etc., within the device.

[0033] 1.3) Data caching and processing preparation: The acquired raw timing data is stored in real-time in on-chip RAM or external cache memory. A predetermined time window (e.g., 200ms from the trailing edge of the trigger signal) or a steady-state judgment algorithm (e.g., the rate of change of current is below a threshold) is used to automatically terminate the acquisition. Subsequently, the system passes the complete timing data block to the integrated edge computing unit and prepares to execute step S2.

[0034] Preferably, the predetermined time window mentioned in step S1 is an adaptive time window, calculated as follows: The short-time energy change rate of the current signal is calculated in real time. When the short-time energy change rate is lower than 1.5 times the background noise energy threshold for K consecutive power frequency cycles, the start-up transient is determined to have ended, and the data acquisition window is dynamically truncated as a predetermined time window.

[0035] To calculate the size of the adaptive time window, the system, while collecting current time-series data, calculates the short-time energy (i.e., the integral of the square of the current) of the current signal in real time within each power frequency cycle (e.g., 20ms). It continuously calculates the rate of change of short-time energy between adjacent power frequency cycles, which directly reflects the degree of drastic change in energy input during equipment startup. The system presets or learns a background noise energy threshold, i.e., the background noise energy level of the socket under no-load or steady-state conditions. The judgment threshold is usually set to 1.5 times this background noise energy to provide a reasonable buffer. When the system detects that the rate of change of short-time energy of the current is lower than the above judgment threshold for K consecutive times, such as 3-5 power frequency cycles, it considers that the equipment startup impact process has ended, the electrical characteristics have been fully displayed, and the system has entered a stable phase. At this time, the system immediately and dynamically truncates data acquisition and defines the time from the moment the equipment is connected to this moment as the "predetermined time window" for this acquisition.

[0036] The startup time varies significantly between different devices; for example, a small controller might take tens of milliseconds, while a large motor might take hundreds of milliseconds. A fixed time window might truncate incomplete features if set too short, or collect a large amount of useless steady-state noise if set too long. An adaptive window ensures complete capture of the unique startup transient process of each device, without omissions or redundancy, providing the highest quality data source for fingerprint recognition. This method allows the socket to intelligently adapt to various unknown and new device models without needing to pre-set different acquisition time parameters for different device categories. The system determines the acquisition duration based on the device's own electrical behavior, resulting in stronger robustness and generalization capabilities. It avoids processing long, meaningless steady-state data, reduces the amount of data to be processed, saves limited local storage and computing resources, and allows edge computing units to focus more on analyzing valuable feature segments.

[0037] In summary, step S1, by locking the relay before power-on and collecting signals only in the test circuit, completely eliminates any possibility of illegal or faulty devices obtaining main power energy during the identification process, physically isolating the risk outside the system. The transient electrical response of the device during startup (such as surge shape and harmonic components) directly reflects its internal circuit topology, motor characteristics, filter capacitors, and other parameters, exhibiting high uniqueness and stability. The data collected during this window provides high-quality, high-information-density raw materials for subsequent high-precision device fingerprint recognition. By strictly limiting the data to be analyzed within an extremely short time window, the amount of data is significantly compressed, making real-time feature extraction and model inference possible on the limited computing and storage resources of the socket, a prerequisite for achieving "millisecond-level" response. The entire detection and data preparation process is completed entirely within the socket, without relying on any network communication, ensuring the system's offline operation capability.

[0038] In one embodiment, step S2 is implemented as follows: First, the raw voltage and current time-series data collected in step S1 are subjected to power frequency filtering, denoising, and normalization to eliminate the influence of power grid harmonics and random interference, ensuring the stability of feature extraction. After data preprocessing, the edge computing unit performs the following core feature calculations in parallel or serially to construct a comprehensive feature vector. The extracted multidimensional features include: Peak inrush current at startup: The maximum value is found directly in the pre-processed current sequence. This is the most intuitive characteristic characterizing the intensity of the equipment startup impact.

[0039] Energy entropy of the current rising edge: First, extract the "rising edge" segment from the start of the current rise to its peak (or reaching steady state). Perform wavelet packet transform on this segment to decompose it into multiple subspaces with different frequency bands. Calculate the energy in each frequency band subspace and the Shannon entropy of the energy distribution of the entire rising edge signal across each frequency band. This entropy value reflects the complexity and regularity of the startup energy release process: stable motors have lower entropy values, while devices with complex switching power supplies have higher entropy values.

[0040] Waveform distance metric based on DTW: The Dynamic Time Warping algorithm is used to compare the similarity of two time series that may differ in length. In practice, the preprocessed entire startup current waveform is compared with each template waveform in a standard device template library pre-stored in local flash memory using DTW. The DTW algorithm "stretches" or "compresses" the time axis to find the optimal matching path between the two sequences and calculates their cumulative distance. This minimum distance is the similarity metric between the current waveform and a given template (the smaller the distance, the more similar). The system typically records the distance to the best-matching template and the distance difference to the most easily confused template as features.

[0041] Furthermore, the scalar values ​​obtained from the above calculations, such as [peak current, energy entropy, DTW distance 1, DTW distance 2, ...], are combined into a fixed-dimensional feature vector. This vector is the "fingerprint" abstraction of the original waveform data, preparing it for subsequent lightweight model input. This embodiment compresses the original waveform, which lasts for hundreds of milliseconds and contains tens of thousands of data points, into a feature vector consisting of only a few to dozens of key numbers. This greatly reduces the computational burden on the subsequent recognition model, making it possible to deploy complex models at resource-constrained edge devices. The multi-dimensional feature fusion constructs a vector that jointly ensures the uniqueness and stability of the "device fingerprint," significantly improving the accuracy and anti-interference capability of the recognition.

[0042] In one embodiment, the core of step S3 is to utilize a lightweight artificial intelligence model at the edge to perform real-time matching and identification of the "device fingerprint" and output structured decision information. Its implementation relies on a software model and algorithm embedded in local memory, and the specific process is as follows: Model loading and initialization: After the system is powered on, the lightweight device fingerprint recognition model and its standard template library (authorized device feature library) pre-installed in the socket Flash or EEPROM are loaded into the memory of the edge computing unit. This model is usually a miniature classifier that has been pruned, quantized, or designed with a special architecture (neural network suitable for MCU), such as a small fully connected network, support vector machine (SVM), or random forest.

[0043] In a preferred embodiment, the lightweight device fingerprint recognition model is constructed by paralleling a Convolutional Neural Network (CNN) branch and a Gradient Boosting Decision Tree (GBDT) branch. These branches extract waveform morphological features and statistical features respectively before fusing them for decision-making. The CNN branch specializes in waveform morphological feature extraction, while the GBDT branch excels at handling structured statistical features. By constructing a parallel lightweight hybrid model, the computational load is controllable, and the parallel computing power is close to the slower of the two branches, rather than the other, thus improving computational efficiency. In the hybrid model, the CNN has a certain tolerance for local waveform deformation and small time shifts; the GBDT is insensitive to small fluctuations in feature values. Since both branches make judgments from different perspectives, even if the judgment accuracy of one branch decreases due to noise interference, the other branch may still provide correct information. The overall decision after fusion is more stable and reliable, achieving a balance between recognition accuracy, inference speed, and system robustness.

[0044] Forward inference computation: The multi-dimensional electrical feature vector obtained in step S2 is fed into the loaded lightweight model as input. The model performs one forward propagation or decision computation, the core task of which is to perform multi-class classification or similarity matching. The output layer typically provides two key pieces of information: Equipment Type Identifier (ID): The most likely equipment category calculated by the model (e.g., "Inverter of Model A", "Welding Robot No. 203").

[0045] Confidence score: A numerical value between 0 and 1 (or a percentage) that represents how certain the model is about the current classification result. This is usually derived from the probability output of the Softmax layer or a similarity score based on a distance metric.

[0046] Related Data Indexing and Output: A device information database is maintained in local storage, which corresponds one-to-one with the model's standard template library. Each record contains at least: device ID, preset safe operating parameters, including maximum allowable power, normal current range, historical energy consumption baseline, or typical operating mode. After the model identifies the device ID, the system immediately indexes and reads the associated historical operating data from the local database based on this ID, such as the average power over the past 24 hours, the duration of the last operation, and safety policy parameters. Finally, the output of step S3 is a structured decision package: {device type identifier, confidence score, associated historical data / policy parameters}.

[0047] Since all identification calculations in step S3 are performed locally on the socket, without relying on any network connection or cloud server, the device authentication process is completely immune to network latency, interruptions, or cloud service failures. This ensures the absolute reliability and real-time performance of the core security mechanism of "identify first, then power on" under any network conditions. Furthermore, the output confidence score is crucial for intelligent decision-making. It enables the system not only to make binary "yes or no" judgments but also to assess the reliability of those judgments. This provides a scientific and quantitative basis for the more refined dynamic access verification in subsequent step S4, enhancing the system's flexibility and fault tolerance.

[0048] In one embodiment, step S4, which involves performing a dynamic admission verification process based on the confidence score and a preset confidence interval, includes: Step S41: When the confidence score is greater than or equal to the high confidence threshold, the direct conduction mode is entered; the device type identifier is checked to see if it is in the whitelist of the current socket's access policy, and the operating power is assessed to see if it does not exceed the current socket's remaining capacity threshold; if the verification is successful, a conduction command is generated to control the relay to close, and the overcurrent protection threshold is dynamically configured according to the historical operating data associated with the device type. Because the model output confidence score is extremely high (e.g., ≥ 90%), the device's identity is almost certain. The system immediately verifies whether the identified device type identifier exists in the socket's pre-set access policy whitelist. Combining the typical power or instantaneously collected power data of this device type, it assesses whether it does not exceed the remaining capacity threshold of the current socket circuit or bus. If all the above verifications pass, while generating a conduction command to control the relay to close, the system retrieves the device's normal operating current peak value from its historical operating data and dynamically and adaptively sets an overcurrent protection threshold slightly higher than this peak value, replacing the uniform fixed threshold. In this way, the authorized device receives power and enjoys precise overcurrent protection "tailor-made" for it.

[0049] Step S42: When the confidence score is greater than or equal to the low confidence threshold and less than the high confidence threshold, enter the active detection mode: control the relay to perform a pulse switching operation of a preset duration, collect the back electromotive force characteristics of the device at the moment of pulse cut-off and the residual charge release characteristics at the second connection, update the multidimensional electrical feature vector, and return to execute step S3. The confidence score is in the moderate range (e.g., 60% ~ 89%), meaning the model cannot be completely certain, but the probability that the device is of a known type remains high. Instead of making a final decision immediately, the system controls the relay to perform a precise pulse-based on / off operation (e.g., closing for 50 milliseconds and then immediately opening). During this process, two key new features are simultaneously acquired: 1) the back electromotive force generated by the device's inductive element at the moment of pulse cut-off; and 2) the impact feature of residual charge release from the device's capacitor at the moment of re-connection. These new features are fused with the initial features to update the multidimensional electrical feature vector, and this enhanced feature vector is returned to step S3 for a second, more accurate identification.

[0050] Preferably, the pulse duration of the pulse-type on / off operation in the active detection mode is calculated as follows: ; In the formula, The duration of the pulse. This is an estimated value for the load resistive component. For safety, this is used to ensure that the induced voltage generated at the moment the pulse is disconnected does not exceed the withstand voltage limit of the socket components; This is an estimated value for the inductive component of the load.

[0051] It is understandable that the pulse duration of the on / off operation... This refers to the length of time a control switch (such as a relay or MOSFET) remains ON during the "active detection" phase of the control system. Its function is to allow the current to build up in the inductive load and reach a stable, estimated value. If the time is too short, the current will not reach the expected value, resulting in inaccurate detection data. If the time is too long, for large inductive loads, the current may be excessive, leading to overheating or triggering overcurrent protection; or, upon disconnection, an extremely high reverse voltage spike may occur.

[0052] In the above formula, This is an estimated value for the inductive component of the load, i.e., the equivalent inductance of the load under test. This is an estimate, usually not measured directly, but calculated through previous probing steps. In the initial stage of active probing, the system may apply a very short test pulse to measure the rate of current rise. Given the applied voltage and the measured rate of current change, this value is approximately equal to the ratio of the applied voltage to the rate of current change. This is an estimate of the load resistive component. Typically, when the current tends to stabilize, the inductor is equivalent to a short circuit. The magnitude is approximately equal to the ratio of the applied voltage to the steady-state current. This is a dimensionless correction factor used to provide a safety margin. When the switch is opened, the inductive load generates a back electromotive force. The rate of change of current at the instant of disconnection is proportional to the magnitude of the current before disconnection, and the magnitude of the current before disconnection is related to the conduction time. This is relevant. Therefore, the formula actually limits the maximum current at the instant of disconnection, thereby limiting the back electromotive force. The value depends on the ratio of the rated voltage of the socket components to the actual operating voltage. For example, if the system operates at 220V and the switch withstand voltage is 600V, the theoretical maximum allowable overshoot is approximately 380V. Engineers will set this value based on experimental data. The value is typically between 0.5 and 0.9, but needs to be calibrated experimentally. Assuming the smart socket detects a load with a mains voltage of 220V, an estimated resistance of 100Ω, an estimated inductance of 0.5H, a time constant of 5ms, and a safety factor of 0.8, the pulse duration can be calculated using the formula above. When the system actively detects the load, it should control the switch to turn on for 4 milliseconds and then turn off. This ensures that at the moment of disconnection, the current rises to approximately 55% of the steady-state value, and the induced voltage generated at this time will not damage the semiconductor switch inside the socket.

[0053] Step S43: When the confidence score is less than the low confidence threshold, or when step S42 fails the verification, a rejection command is generated to keep the relay open and an alarm message is output.

[0054] When the confidence score is too low (e.g., <60%), or when the re-identification in step S42 still fails to reach a high confidence level, the system strictly keeps the relay in the open state and generates a rejection command. Simultaneously, alarm information is output via local indicator lights (e.g., flashing red light) and communication interfaces (e.g., sending alarms to the platform), and an event log is recorded. This means that potentially unauthorized or faulty devices are absolutely prevented from powering on, and management personnel are notified to intervene.

[0055] This system avoids a "one-size-fits-all" approach. For clearly legitimate devices (high confidence), authorization is granted rapidly, minimizing waiting time. For suspected legitimate devices (medium confidence), secondary confirmation is performed through security probes, preventing false rejection of legitimate devices due to momentary interference and improving system availability. For clearly illegitimate or unidentifiable devices (low confidence), they are resolutely blocked, ensuring the highest level of security. The "active detection mode" transforms the system from a "passive observer" to an "active interactor." Back electromotive force and residual charge release characteristics deeply reflect the physical properties of devices, providing strong discriminative power. This mechanism effectively addresses complex scenarios such as varying device temperatures and slight variations in initial startup characteristics, significantly improving the fault tolerance and accuracy of identification in complex edge environments. Furthermore, the system not only controls on / off states but also dynamically configures overcurrent protection thresholds based on historical device data. This avoids both overly wide threshold settings leading to protection failure and overly narrow threshold settings leading to false tripping, achieving dual protection for both safety and stable operation.

[0056] In one embodiment, the admission strategy in step S41 further includes a thermal-electrical coupling security verification sub-step: Obtain the real-time temperature and historical temperature rise curve slope of key internal nodes of the socket; Based on the rated root mean square current corresponding to the identified equipment type and the estimated contact resistance of the current relay contact, a short-time temperature rise prediction model is constructed to obtain the predicted temperature value. The predicted temperature value is compared with the tolerance limit temperature of the insulation material. When the predicted temperature value exceeds the tolerance limit temperature of the insulation material, a rejection command is generated directly, and the socket is locked until the predicted temperature value drops below the tolerance limit temperature.

[0057] This embodiment aims to intelligently correlate real-time temperature status with impending electrical behavior, achieving proactive protection. The system not only monitors the current after power is applied but also continuously "senses" its own temperature before power is supplied. Through built-in temperature sensors, it reads the current temperature of key "heat-generating points" such as relay contacts and terminals in real time and analyzes their recent trends (temperature rise slope) to understand the socket's "health baseline" and "heating trend." Upon identifying the type of device to be connected, the system obtains the device's rated current parameters. Simultaneously, based on the relay's operating time and electrical performance, it estimates its current contact resistance (increased contact resistance leads to abnormal heating). The system inputs this information, including device current, contact resistance, current temperature, and temperature rise trend, into a built-in simplified thermodynamic model. This model quickly simulates how high the temperature of key parts of the socket will rise in the short term if the device is powered at that moment, thus obtaining a scientifically predicted temperature value. The system does not wait for danger to occur. It compares this predicted temperature value with the maximum safe tolerance temperature of the socket's insulation material. If the forecast indicates that the power supply is about to exceed the safety threshold, the system will decisively refuse power before it is applied. This is equivalent to anticipating and preventing risks before an accident occurs. Once the power is refused due to predicted overheating, the socket will not simply disconnect; instead, it will enter a "safety lockout" state, preventing any new connection attempts. It will continuously monitor the temperature drop, and will only automatically unlock and restore service after the predictive model confirms that the temperature has stabilized and dropped to an absolutely safe range. This forms a self-protective intelligent feedback loop.

[0058] Thus, this embodiment changes the logic of traditional overload protection, shifting from reactive "tripping and remediation" to proactive "predictive interception." It no longer protects against obvious instantaneous high-current overloads, but rather against hidden, slowly developing thermal hazards, such as "chronic heating" caused by aging contacts and poor heat dissipation due to dust accumulation, thereby nipping electrical fires in the bud. The protection threshold is no longer a fixed, static value at the factory, but rather dynamic data calculated based on the specific equipment current, the specific socket status (resistance, current temperature), and the real-time environment (temperature rise trend), making the protection action more precise. The entire prediction, decision-making, locking, and recovery process is completed autonomously at the socket edge, without relying on any external commands, greatly improving the reliability and independence of the entire system.

[0059] In one embodiment, the method further includes: The control edge computing unit continuously records the electrical characteristic vectors of the actual operation phase during the device operation process; Calculate the Mahalanobis distance between the electrical feature vectors during actual operation and the cluster centers of the equipment types stored locally; If the Mahalanobis distance exceeds the dynamic threshold that decays over time for N consecutive times, it is determined that the device status has drifted or the model has changed; the automatic access function of the socket is suspended, marked as "pending verification", and the abnormal fragment is encrypted and uploaded to the host computer, requesting manual or cloud-based large model intervention for recalibration.

[0060] After the device is powered on and operating normally, the system does not stop working. It continuously analyzes the current and voltage characteristics of the device during operation and calculates in real time the degree of deviation (Mahavira distance) between these characteristics and the device's "standard health record" (cluster center). This deviation alarm threshold is not fixed but gradually widens as the device operates normally to avoid false alarms. Once the system repeatedly detects a significant and persistent deviation between the current operating status and the "health record," it will determine that the device may be aging, have a hidden fault, or have been replaced without authorization. At this time, the system will immediately suspend the automatic identification and access function of the socket, switching to a state requiring manual verification. At the same time, it will encrypt this abnormal data and send it to the backend administrator or a more powerful cloud analysis system for final diagnosis.

[0061] This embodiment adds a continuous "health check" step to the daily operation of the equipment. It can keenly detect the slow deterioration of the equipment's condition or unauthorized changes, provide early warnings before potential failures cause accidents, and submit the problem to a higher level of intelligence for processing, thereby realizing an intelligent upgrade from "one-time authentication upon access" to "full lifecycle reliability management".

[0062] See Figure 2 In one embodiment, the present invention also provides an industrial smart socket based on edge computing, comprising: The power conversion module is used to provide operating power to the internal circuitry of the socket; The plug detection module is used to detect the plug insertion status and generate a trigger signal when the plug insertion is detected; Relays and drive circuits are used to control the on / off state of sockets; Current and voltage sampling circuits are used to collect electrical characteristic data of connected devices; An edge computing unit is connected to the plug detection module, the current and voltage sampling circuit, and the relay drive circuit. The edge computing unit includes a microcontroller and an internally integrated non-volatile memory. The non-volatile memory is pre-loaded with a lightweight device fingerprint recognition model and an access strategy library. The communication module is used for data interaction with the host computer; The edge computing unit is configured to perform a method as described in the first aspect above and any possible implementation thereof.

[0063] Preferably, the industrial smart socket based on edge computing further includes a local alarm module connected to the edge computing unit; the local alarm module includes one or more of an indicator light, a buzzer, or a display screen, and is used to output local alarm information when the device fails to be recognized.

[0064] Preferably, in the edge computing-based industrial smart socket, an analog-to-digital converter is provided between the edge computing unit and the current and voltage sampling circuit, and the sampling rate of the analog-to-digital converter is not less than 10kHz.

[0065] It is understood that the functions or modules of the industrial smart socket provided in this embodiment can be used to execute the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0066] In one embodiment, the present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, the computer program including program instructions; when the program instructions are executed by the edge computing unit, the edge computing unit performs the edge computing-based industrial smart socket safety and energy-saving control method as described in any of the foregoing embodiments.

[0067] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A safety and energy-saving control method for industrial smart sockets based on edge computing, characterized in that, The method includes: Step S1: In response to detecting that a plug is inserted into a socket or receiving a trigger signal for device access, while keeping the control relay open, collect timing data of voltage and current of the accessed device during the startup transient period, wherein the startup transient period is a predetermined time window from the moment the device is accessed until the device enters steady-state operation. Step S2: Based on the edge computing unit integrated inside the socket, extract multi-dimensional electrical feature vectors from the time-series data. The multi-dimensional electrical feature vectors include at least the peak value of the initiation surge current, the energy entropy of the current rising edge, and the distance measure between the current waveform calculated based on the DTW algorithm and the standard template library. Step S3: Input the multi-dimensional electrical feature vector into the locally pre-stored lightweight device fingerprint recognition model, and output the device type identifier, historical operating data associated with the device type, and the corresponding confidence score; Step S4: Perform a dynamic admission verification process based on the confidence score and the preset confidence threshold.

2. The safety and energy-saving control method for industrial smart sockets based on edge computing according to claim 1, characterized in that, Step S4, which describes performing a dynamic admission verification process based on the confidence score and a preset confidence interval, includes: Step S41: When the confidence score is greater than or equal to the high confidence threshold, the direct conduction mode is entered; the device type identifier is checked to see if it is in the whitelist of the current socket's access policy, and the operating power is assessed to see if it does not exceed the current socket's remaining capacity threshold; if the verification is successful, a conduction command is generated to control the relay to close, and the overcurrent protection threshold is dynamically configured according to the historical operating data associated with the device type. Step S42: When the confidence score is greater than or equal to the low confidence threshold and less than the high confidence threshold, enter the active detection mode: control the relay to perform a pulse switching operation of a preset duration, collect the back electromotive force characteristics of the device at the moment of pulse cut-off and the residual charge release characteristics at the second connection, update the multidimensional electrical feature vector, and return to execute step S3. Step S43: When the confidence score is less than the low confidence threshold, or when step S42 fails the verification, a rejection command is generated to keep the relay open and an alarm message is output.

3. The safety and energy-saving control method for industrial smart sockets based on edge computing according to claim 2, characterized in that, The admission strategy described in step S41 further includes a thermal-electric coupling security verification sub-step: Obtain the real-time temperature and historical temperature rise curve slope of key internal nodes of the socket; Based on the rated root mean square current corresponding to the identified equipment type and the estimated contact resistance of the current relay contact, a short-time temperature rise prediction model is constructed to obtain the predicted temperature value. The predicted temperature value is compared with the tolerance limit temperature of the insulation material. When the predicted temperature value exceeds the tolerance limit temperature of the insulation material, a rejection command is generated directly, and the socket is locked until the predicted temperature value drops below the tolerance limit temperature.

4. The safety and energy-saving control method for industrial smart sockets based on edge computing according to claim 1, characterized in that, The lightweight device fingerprint recognition model described in step S3 is composed of a convolutional neural network branch and a gradient boosting decision tree branch connected in parallel. After extracting waveform morphology features and statistical features respectively, the models are fused together for decision.

5. The safety and energy-saving control method for industrial smart sockets based on edge computing according to claim 1, characterized in that, The predetermined time window mentioned in step S1 is an adaptive time window, calculated as follows: The short-time energy change rate of the current signal is calculated in real time. When the short-time energy change rate is lower than 1.5 times the background noise energy threshold for K consecutive power frequency cycles, the start-up transient is determined to have ended, and the data acquisition window is dynamically truncated as a predetermined time window.

6. The safety and energy-saving control method for industrial smart sockets based on edge computing according to claim 1, characterized in that, The method further includes: The control edge computing unit continuously records the electrical characteristic vectors of the actual operation phase during the device operation process; Calculate the Mahalanobis distance between the electrical feature vectors during actual operation and the cluster centers of the equipment types stored locally; If the Mahalanobis distance exceeds the dynamic threshold that decays over time for N consecutive times, it is determined that the device status has drifted or the model has changed; the automatic access function of the socket is suspended, marked as "pending verification", and the abnormal fragment is encrypted and uploaded to the host computer, requesting manual or cloud-based large model intervention for recalibration.

7. An industrial smart socket based on edge computing, characterized in that, include: The power conversion module is used to provide operating power to the internal circuitry of the socket; The plug detection module is used to detect the plug insertion status and generate a trigger signal when the plug insertion is detected; Relays and drive circuits are used to control the on / off state of sockets; Current and voltage sampling circuits are used to collect electrical characteristic data of connected devices; An edge computing unit is connected to the plug detection module, the current and voltage sampling circuit, and the relay drive circuit. The edge computing unit includes a microcontroller and an internally integrated non-volatile memory. The non-volatile memory is pre-loaded with a lightweight device fingerprint recognition model and an access strategy library. The communication module is used for data interaction with the host computer; The edge computing unit is configured to execute the edge computing-based industrial smart socket safety and energy-saving control method as described in any one of claims 1 to 6.

8. The industrial smart socket based on edge computing according to claim 7, characterized in that, It also includes a local alarm module connected to the edge computing unit; the local alarm module includes one or more of an indicator light, a buzzer or a display screen, for outputting local alarm information when the device fails to recognize it.

9. The industrial smart socket based on edge computing according to claim 7, characterized in that, An analog-to-digital converter is provided between the edge computing unit and the current and voltage sampling circuit, and the sampling rate of the analog-to-digital converter is not less than 10kHz.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions; when executed by the edge computing unit, the program instructions cause the edge computing unit to perform the edge computing-based industrial smart socket safety and energy-saving control method as described in any one of claims 1 to 6.

Images