A real-time current monitoring method and system applied to an anti-overload power line

Abstract

The application relates to the technical field of power safety, in particular to a real-time current monitoring method and system applied to an overload-preventing power line, which comprises the following steps: collecting current waveform data of a load on the power line in real time at a preset sampling frequency; after a new load is monitored to be connected and stably operated, a learning period is triggered, and multi-dimensional feature parameters of the current waveform data are extracted in the learning period; the multi-dimensional feature parameters are combined to generate a load digital fingerprint representing a normal operation mode of the load, and the load digital fingerprint is stored in a nonvolatile memory as a dynamic safety baseline; after the learning period ends, real-time current waveform data is continuously collected, and corresponding real-time multi-dimensional feature parameters are calculated; the real-time multi-dimensional feature parameters are compared with the stored dynamic safety baseline, and are determined to be abnormal when at least one of the following preset logics is met. The application solves the core technical problem that normal surges and dangerous overloads cannot be effectively distinguished in the prior art.

Description

Technical Field

[0001] This application relates to the field of power safety technology, and in particular to a real-time current monitoring method and system for overload protection power lines. Background Technology

[0002] In modern home and office environments, with the rapid increase in the number and types of electrical equipment, higher requirements are placed on circuit overload protection. Traditional overload protection devices, such as fuses and thermal-magnetic circuit breakers, mainly rely on the thermal or magnetic effects generated when the current exceeds the rated value to trigger protection. Their response mechanism is relatively simple and cannot accurately distinguish between the normal instantaneous large current generated when inductive loads such as motors start and the continuous dangerous overload caused by line or equipment faults. This often leads to unnecessary tripping or failure to provide timely warnings in the early stages of danger.

[0003] To address the aforementioned issues, some digital monitoring-based solutions have emerged in the existing technology. However, these technologies still have significant limitations when applied to complex real-world power consumption scenarios. For example, solutions aimed at device identification primarily focus on identifying the device rather than assessing its current operational safety. Consequently, their accuracy and reliability in overload protection applications are insufficient. Furthermore, when multiple different types of loads are connected to a single monitoring point, existing technologies struggle to effectively decompose and independently evaluate the mixed current waveforms. They cannot accurately determine which load has changed or experienced an anomaly, exhibiting poor adaptability to multi-load scenarios. This prevents the aggregation and evaluation of the total load across multiple monitoring points connected to the same physical branch, thus failing to prevent overload risks at the branch level. Summary of the Invention

[0004] The objective of this application is to provide a real-time current monitoring method for overload protection power lines, comprising: acquiring current waveform data of a load on the power line in real time at a preset sampling frequency using a current acquisition device; triggering a learning cycle after detecting a new load connection and stable operation, extracting multi-dimensional feature parameters of the current waveform data within the learning cycle, the multi-dimensional feature parameters including: peak surge current, surge duration, steady-state root mean square current, current crest factor, and harmonic component characteristics; combining the multi-dimensional feature parameters to generate a load digital fingerprint characterizing the normal operation mode of the load, and storing the load digital fingerprint in a non-... In volatile memory, a dynamic safety baseline is used. After the learning cycle ends, real-time current waveform data is continuously collected, and the corresponding real-time multidimensional characteristic parameters are calculated. The real-time multidimensional characteristic parameters are compared with the stored dynamic safety baseline, and an anomaly is determined when at least one of the following preset logic is met: when the real-time steady-state root-mean-square current exceeds a preset percentage of the corresponding steady-state root-mean-square current in the dynamic safety baseline for a preset duration, it is determined to be a slow overload anomaly; when the real-time peak surge current exceeds a preset percentage of the corresponding peak surge current in the dynamic safety baseline, it is determined to be a startup instantaneous anomaly.

[0005] By adopting the above technical solution, the intelligent triggering logic is clearly divided into judgment rules for two different abnormal scenarios. Thus, through a deep understanding of load behavior, the core technical problem of the inability to effectively distinguish between normal surges and dangerous overloads in the existing technology is successfully solved.

[0006] Optionally, when multiple loads are connected to the power line, the method further includes: identifying a power step change by monitoring the rate of change of total power, and defining the power step change as a load change event; when the load change event occurs, extracting the difference between the current waveform data before and after the event to generate a difference digital fingerprint; matching the difference digital fingerprint with a preset general electrical fingerprint database to identify the type of load that has changed; updating an online load status list based on the identification result, the online load status list recording all currently online load types, and dynamically generating a composite safety baseline for overall risk assessment based on the online load status list.

[0007] By adopting the above technical solution, an event-driven lightweight load decomposition concept is introduced, which efficiently solves the problem of adaptability of a single monitoring point to multiple load scenarios.

[0008] Optionally, the general electrical fingerprint database includes feature models for at least three load types: resistive load model, inductive load model, and switching power supply load model.

[0009] By adopting the above technical solutions, the identification process becomes more accurate.

[0010] Optionally, the composite safety baseline is generated by: arithmetically summing the steady-state root mean square current characteristic values ​​of all loads in the online load status list, and combining them with the expected surge current characteristics of the inductive load to calculate the composite current upper limit and waveform characteristics.

[0011] By adopting the above technical solution, the generation logic of the composite safety baseline is specifically explained, which enables overload protection under multi-load environment to have predictive capabilities, greatly improves the intelligence level of the protection logic and the reliability of practical application, and makes the safety assessment under multi-load environment more forward-looking and reliable.

[0012] Optionally, the preset logic for determining an anomaly further includes: when the offset between the real-time harmonic component characteristics and the corresponding harmonic component characteristics in the dynamic safety baseline exceeds a preset threshold, the device is determined to be faulty.

[0013] By adopting the above technical solutions, early warning of potential equipment failures can be achieved, which is a more refined and forward-looking protection method than traditional overcurrent protection.

[0014] Optionally, the method further includes: constructing a wireless mesh network covering a specific physical space using a low-power wireless communication module integrated in the power line monitoring point; each monitoring point in the wireless mesh network periodically broadcasting data packets containing its own load status information to the network; a central node receiving and aggregating load status information from multiple monitoring points within the same circuit group to calculate the total load of the circuit group; comparing the calculated total load with the rated threshold of the circuit corresponding to the circuit group, and executing an early warning or protection command when the total load exceeds the rated threshold of the circuit.

[0015] By adopting the above technical solution and introducing a low-power wireless mesh network, multiple independent monitoring points are connected into a collaborative system, which completely solves the technical pain points of existing technologies where a single monitoring point cannot perceive the global load and cannot provide branch-level protection.

[0016] Optionally, the data packet content of the load status information includes at least: monitoring point device ID, current load power, load type, and current safety margin percentage.

[0017] Optionally, the method further includes: uploading the load digital fingerprint and load change event data to a cloud server in an anonymized manner; the cloud server training the uploaded data using big data analysis and machine learning algorithms to optimize and expand the general appliance fingerprint database; and pushing the updated general appliance fingerprint database to monitoring points in the network via online firmware upgrades.

[0018] By adopting the above technical solutions, the system acquires self-evolution capabilities.

[0019] Optionally, the triggering condition for the learning cycle is: after the current is detected to rise a step from zero or a stable base value, the root mean square value of the current fluctuates less than a preset stable threshold during a preset stable observation period.

[0020] The second objective of this application is to provide a real-time current monitoring system for overload protection power lines, comprising: a processor; and a memory electrically connected to the processor; wherein the memory stores computer-executable instructions, which, when executed by the processor, cause the system to perform the aforementioned method. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the real-time current monitoring method for overload protection power lines applied in this application. Detailed Implementation

[0022] To better understand this application, it will be described in more detail below with reference to embodiments.

[0023] like Figure 1 As shown in the figure, this application discloses a real-time current monitoring method for overload protection power lines, including the following steps.

[0024] S01: The current waveform data of the load on the power line is collected in real time through the current acquisition device at a preset sampling frequency.

[0025] Understandably, the input to this process is the alternating current flowing through the power supply's live wire. This alternating current is an analog signal with a frequency of 50 Hz or 60 Hz and an amplitude that varies with the load. The current acquisition device can employ a high-precision shunt resistor and an analog-to-digital converter integrated with a microcontroller. The high-precision shunt resistor can be a high-stability metal foil resistor with a resistance value set to 1 milliohm, an accuracy better than ±0.1%, and a temperature coefficient less than 20 parts per million per degree Celsius. This high-precision shunt resistor is connected in series with the power supply's live wire and, according to Ohm's law, linearly converts the instantaneous value of the real-time current into a weak voltage signal. The voltage signal is equal to the instantaneous value of the real-time current multiplied by the resistance value of the shunt resistor. This voltage signal is then sent to the analog-to-digital converter, which can be a 16-bit high-precision converter. The analog-to-digital converter (ADC) communicates with the main microcontroller via an internal integrated circuit bus. This microcontroller can be a high-performance microcontroller with a hardware floating-point unit. It incorporates a high-performance 32-bit core operating at 168 MHz. The microcontroller configures the ADC via the internal integrated circuit bus, setting its sampling gain to match the voltage range generated by the shunt resistor and initiating continuous conversion mode. Its preset sampling frequency is set to 3.2 kHz. This sampling frequency is based on the Nyquist sampling theorem, ensuring distortion-free capture of the highest 32nd harmonic components of the 50 Hz power frequency signal, covering the harmonic spectrum range generated by most nonlinear loads. To achieve uninterrupted high-speed data stream acquisition, the microcontroller's internal Direct Memory Access (DMemory) controller can be configured in a dual-buffered circular transfer mode. Without CPU intervention, the DMemory controller automatically transfers the 16-bit digital data generated by the analog-to-digital converter (ADC) from the ADC's data register or the microcontroller's internal integrated circuit bus receive buffer to one of two circular buffers, each with 256 sampling points, within the microcontroller's internal Static Random Access Memory (SRAM). When one buffer is full, the DMemory controller automatically switches to the other buffer to continue writing, simultaneously generating a half-full or full-full interrupt to notify the CPU that the data in the filled buffer can be processed. The output is a series of discrete 16-bit signed integer values ​​stored in SRAM, representing the instantaneous current values ​​over the most recent power frequency cycles. The quantization unit is calibrated to amperes via a calibration procedure, and the range is set based on the shunt resistor and ADC gain, for example, covering a range from -20 amperes to +20 amperes.

[0026] S02: After detecting the connection of a new load and its stable operation, a learning cycle is triggered, and the multi-dimensional feature parameters of the current waveform data are extracted within the learning cycle.

[0027] It is understood that the triggering condition for the learning cycle is: after a significant step increase in current from zero or a stable base value is detected, the root mean square value of the current fluctuates less than a preset stable threshold within a preset stable observation period; the input to this process is the aforementioned real-time current waveform data stream; the processing mechanism is implemented by a state machine algorithm running inside the microcontroller, which continuously calculates and monitors the real-time effective value of the current (i.e., the root mean square value); the real-time effective value of the current is obtained by averaging the sum of the squares of several current sample values ​​from the most recent power frequency cycle and then taking the square root. For example, at a sampling rate of 3.2 kHz, 50 Hz corresponds to 64 sampling points. The state machine includes two states: an idle baseline state and a transient observation state. The initial state is the idle baseline state. In this state, the real-time effective current value calculated by the system is lower than a preset no-load or standby current threshold. This no-load or standby current threshold can be set to 50 mA to distinguish between no-load state and small standby power consumption. At this time, the system continuously monitors whether the real-time effective current value exceeds a significant step rise threshold. This significant step rise threshold can be set to 250 mA to effectively identify the clear action of a new appliance starting up, while filtering out small current changes caused by power supply voltage fluctuations or internal state switching of existing equipment. Once the real-time effective current value jumps from below the no-load or standby current threshold to above the significant step rise threshold, the state machine immediately switches to the transient observation state. In the transient observation state, the system enters a preset stable observation period, the duration of which can be set to 3 seconds, to provide power to most inductive loads such as compressors or switching power supplies. The source completes the startup transient process and enters a stable operating mode within a sufficient time. During the stable observation period, the microcontroller calculates the average and standard deviation of the current effective value per second, and determines whether the volatility within that second is less than a preset stability threshold. This stability threshold can be set to 5%, where volatility refers to the standard deviation of the effective value divided by the average effective value. This stability threshold is used to confirm that the load has entered a stable operating state, rather than a state of periodic power fluctuation, such as the operating state of a variable frequency air conditioner. If the volatility within three consecutive 1-second time windows is less than the stability threshold, the stable operating condition is determined to be met, and the output result is that a Boolean learning start flag is set to true inside the microcontroller. This learning start flag will directly trigger the subsequent multi-dimensional feature parameter extraction process. If the current falls back below the threshold or the volatility exceeds the standard during the observation period, the state machine will reset to the idle baseline state, thereby ensuring the reliability of the learning process and the data quality.

[0028] Understandably, when the learning start flag is true, the microcontroller immediately begins a learning cycle of a preset duration, which can be set to 5 seconds. The input to this process is a total of 16,000 current sampling data blocks continuously acquired and stored by the direct memory access controller within those 5 seconds. The processing mechanism involves the microcontroller calling a series of digital signal processing algorithms pre-installed in the firmware to perform parallel or serial analysis on these data blocks in order to extract multi-dimensional feature parameters, i.e., multi-dimensional feature parameters. These multi-dimensional feature parameters include peak surge current, surge duration, steady-state root mean square current, current crest factor, and harmonic component characteristics.

[0029] Specifically, the microcontroller iterates through the first sub-window of the learning cycle data block. The duration of this first sub-window can be defined as a preset surge time window and set to 500 milliseconds. This time window is sufficient to cover the startup surge process of most electrical appliances. Within this time window, the microcontroller searches for the maximum absolute value of all sampling points, and this maximum value is recorded. The output peak surge current is a 32-bit floating-point number in amperes. Its reasonable value range can be obtained based on the rated current of the line, for example, 0 to 50 amperes.

[0030] Understandably, the microcontroller calculates the root mean square (RMS) value of the current data in the latter half of the learning cycle (e.g., from the 3rd to the 5th second) to obtain the steady-state RMS current. Then, it defines a surge discrimination threshold, which is equal to the steady-state RMS current multiplied by 1.5. This 150% coefficient is used to distinguish between steady-state fluctuations and actual startup surges. The microcontroller scans from the beginning of the learning cycle, recording the time when the absolute value of the instantaneous current first exceeds the surge discrimination threshold and the time when it last falls back below the surge discrimination threshold. The difference between the two is the surge duration. The output surge duration is a 32-bit unsigned integer in milliseconds, and its reasonable value range is generally 0 to 5000 milliseconds.

[0031] The microcontroller selects the latter half of the data in the learning cycle that is determined to be stable, such as 6400 sampling points from the 3rd to the 5th second. It applies the above-mentioned root mean square calculation formula to this data segment and averages the calculated root mean square values ​​of multiple power frequency cycles to eliminate short-term fluctuations. The output steady-state root mean square current is a 32-bit floating-point number in amperes, and its reasonable value range is generally 0.05 to 16 amperes.

[0032] Within the same steady-state data segment used to calculate the steady-state root-mean-square current, the microcontroller finds the maximum absolute value of the instantaneous current, i.e., the steady-state peak current. Then, it calculates the current crest factor using a formula that is the steady-state peak current divided by the steady-state root-mean-square current. The output current crest factor is a 32-bit floating-point number, dimensionless. For purely resistive loads, its theoretical value is generally 1.414, and for nonlinear loads such as switching power supplies, it can reach over 3.0.

[0033] The microcontroller extracts a sequence of 1024 sampling points from the steady-state data segment. This length is typically a power of 2 to facilitate the execution of the Fast Fourier Transform (FFT) algorithm. The microcontroller calls an optimized radix-2 FFT function from a digital signal processing library optimized for the microcontroller kernel to transform the sampling point sequence, obtaining a complex array containing the amplitude and phase of each frequency component. Subsequently, the microcontroller extracts the fundamental frequency and the amplitudes of its odd harmonics from this complex array, such as the 3rd, 5th, 7th, and up to the 25th harmonics, and normalizes these amplitudes relative to the fundamental frequency amplitude to form a feature vector. The output harmonic component feature is a one-dimensional array consisting of 12 32-bit floating-point numbers, with the floating-point numbers typically ranging from 0.0 to 1.0.

[0034] S03: Combine the multi-dimensional feature parameters to generate a load digital fingerprint that characterizes the normal operation mode of the load, and store the load digital fingerprint in a non-volatile memory as a dynamic security baseline.

[0035] Understandably, the input to this process consists of the aforementioned multidimensional characteristic parameters: peak surge current, surge duration, steady-state root-mean-square current, current crest factor, and harmonic component characteristics. The processing mechanism involves defining a data structure in the microcontroller's firmware to encapsulate the load digital fingerprint information. This data structure contains member variables that correspond one-to-one with the aforementioned multidimensional characteristic parameters. The microcontroller fills each calculated characteristic value into an instance of this data structure. This data structure instance filled with data constitutes the load digital fingerprint and also serves as the dynamic safety baseline for subsequent anomaly detection. The output is that the microcontroller, through the serial peripheral interface, calls the driver to write this load digital fingerprint data structure instance, along with an internally generated unique timestamp or serial number identifier, into an external non-volatile memory chip, such as a high-capacity serial flash memory chip. This data is stored in a pre-planned specific sector, thereby achieving persistent data storage and ensuring that the dynamic safety baseline is not lost after the device is powered off.

[0036] S04: After the learning cycle ends, real-time current waveform data is continuously collected, and the corresponding real-time multidimensional feature parameters are calculated.

[0037] Understandably, after the learning cycle successfully ends and a dynamic safety baseline is generated, the system enters a continuous real-time monitoring mode. The input to this process is the real-time acquired current waveform data stream and the dynamic safety baseline instance corresponding to the current load state read from non-volatile memory. The processing mechanism is that the microcontroller continuously analyzes the real-time data at a very high frequency in a main loop and compares it with the corresponding dynamic safety baseline. An anomaly is determined when at least one of the following preset logics is met: when the real-time steady-state root-mean-square current continuously exceeds a preset percentage of the corresponding steady-state root-mean-square current in the dynamic safety baseline for a preset duration, it is determined to be a slow overload anomaly; when the real-time peak surge current exceeds a preset percentage of the corresponding peak surge current in the dynamic safety baseline, it is determined to be a startup instantaneous anomaly.

[0038] Specifically, the microcontroller continuously calculates the real-time root-mean-square (RMS) current using a sliding window. The calculation window can be set to 200 milliseconds, or 10 power frequency cycles, to smooth out instantaneous fluctuations. Then, the real-time RMS current is compared with the steady-state RMS current in the dynamic safety baseline. The judgment condition is that the real-time RMS current exceeds the sum of the baseline steady-state RMS current and a preset percentage margin, which can be set to 15%. To prevent false alarms triggered by brief and harmless current overshoot, the judgment logic can also include a time dimension, meaning that the condition needs to be continuously met for a preset duration, which can be set to 10 seconds. An internal hardware timer of the microcontroller starts when the condition is first met. If the condition is not met again during the period, the timer is reset. Only when the timer successfully counts for 10 seconds is the anomaly finally confirmed. The output result can be that a status flag bit inside the system is set to a slow overload abnormal state, thereby triggering a buzzer to emit a continuous alarm sound or cutting off the power supply through a relay.

[0039] Understandably, the logic for determining instantaneous startup anomalies is activated only for a short period after the system identifies a new load startup event through the aforementioned step detection algorithm, such as within 1 second after startup. During this period, the microcontroller calculates the instantaneous peak current in real time and compares it with the peak surge current in the dynamic safety baseline. The determination condition is that the real-time peak current exceeds the sum of the baseline peak surge current and a preset percentage margin of 20%. This percentage margin allows for a certain margin for normal startup surges caused by factors such as grid voltage fluctuations, but it can capture extreme surge currents that far exceed the normal range caused by faults such as motor stall or startup capacitor failure. This is an instantaneous judgment. Once the condition is met, the anomaly can be confirmed without waiting. The output result is that the system's internal status flag is set to the instantaneous startup anomaly state, and protection actions are immediately executed. Its response speed is much faster than slow overload protection, thus effectively protecting the equipment from damage caused by startup failures.

[0040] In this embodiment of the application, when multiple loads are connected to the power line, the method further includes: identifying a step change in power by continuously monitoring the rate of change of total power, and defining the change as a load change event; when the load change event occurs, extracting the difference between the current waveform data before and after the event to generate a difference digital fingerprint; matching the difference digital fingerprint with a pre-set general electrical fingerprint library in the firmware to identify the type of load that has changed; updating an online load status list that records all currently online load types based on the identification result, and dynamically generating a new composite safety baseline for overall risk assessment based on the list.

[0041] Specifically, power step changes can be identified by continuously monitoring the rate of change of total power. In addition to real-time current waveform data, this process also requires synchronously acquired real-time voltage waveform data. This can be achieved by adding a voltage sampling channel based on a resistor divider network and an isolation amplifier within the monitoring unit. This voltage sampling channel operates synchronously with the current sampling channel. The microcontroller uses a hardware floating-point unit to calculate the effective power in real time. The calculation formula is the average of the products of several voltage and current sampling values ​​within one power frequency cycle. Subsequently, the microcontroller calculates the rate of change of power at a fixed time interval, such as 100 milliseconds. The microcontroller then converts the calculated power... The absolute value of the rate of change is compared with a preset power step threshold, which can be set to 30 watts per second. This allows for sensitive detection of events as small as the on or off of an LED light, while ignoring gradual power changes caused by normal power regulation of large appliances, such as inverter air conditioners. When the absolute value of the power rate of change exceeds the power step threshold, it indicates that a load change event has occurred. The output result is that the microcontroller stores two synchronous voltage and current waveform data segments before the event and after the event and the power has stabilized again into waveform data buffers before and after the event, for example, 200 milliseconds before the event and 200 milliseconds after waiting 2 seconds after the event.

[0042] Then, the difference between the current waveform data before and after the event is extracted to generate a difference digital fingerprint. The input to this process is the waveform data buffers before and after the event. The processing mechanism is to perform subtraction in the frequency domain to obtain a more robust result. The microcontroller first performs Fast Fourier Transform on the current data in the buffer before the event and the current data in the buffer after the event, respectively, to obtain two complex spectra. The microcontroller then performs point-by-point complex subtraction on these two spectra in the frequency domain to obtain the difference spectrum, thereby eliminating the electrical characteristics of the original load in the background, so that the difference spectrum contains almost only the newly changed one. The electrical characteristics of the load are analyzed. Based on this difference spectrum, the microcontroller converts it back to a time-domain difference signal using an inverse fast Fourier transform, or directly calculates the same multidimensional characteristic parameters as in the above embodiment in the frequency domain. For example, the steady-state root-mean-square current of the changing load is calculated by analyzing the DC component and harmonic components of the difference spectrum, and the peak surge current and surge duration are extracted by analyzing the difference waveform during the transient phase. The output is a difference digital fingerprint with the same structure as the load fingerprint data, which describes the electrical characteristics of the load that was just turned on or off.

[0043] Next, the differential digital fingerprint is matched with a pre-installed general-purpose electrical fingerprint database in the firmware. The input to this process is the generated differential digital fingerprint and the general-purpose electrical fingerprint database stored in non-volatile memory. The processing mechanism is to execute a pattern matching algorithm in a multi-dimensional feature space, such as a nearest neighbor classification algorithm. The fingerprint database pre-stores feature parameters of various typical load models. The general-purpose electrical fingerprint database contains feature models of at least three load types: resistive load model, inductive load model, and switching power supply load model. Among them, the resistive load model is characterized by a peak factor in the range of [1.40, ...]. In the range of 1.45, the total harmonic distortion is less than 5%, and the ratio of surge current to steady-state current is close to 1.0. The characteristics of the inductive load model are: there is a significant surge current, the peak value of which is usually 5 to 10 times the steady-state root mean square current, the surge duration is between 100 ms and 2000 ms, and the current phase lags behind the voltage. The characteristics of the switching power supply load model are: the crest factor is usually greater than 2.0, the total harmonic distortion exceeds 60%, and the harmonic components are mainly concentrated in low-order odd harmonics. The microcontroller calculates the weighted Euclidean distance between the difference digital fingerprint to be matched and the center point of each model in the GE fingerprint library. The calculation method is to multiply the square of the difference of each feature value by the corresponding weight and then sum and take the square root. Among them, features with high discrimination are given higher weights. The microcontroller finds the library model with the smallest distance and checks whether the smallest distance is less than the preset matching confidence threshold. The output result is that if the match is successful, the type of variable load is identified, such as the inductive load type.

[0044] Finally, the online load status list is updated, and a composite safety baseline is dynamically generated. The input to this process is the identified load type and its difference digital fingerprint, as well as a dynamic array or linked list maintained in the microcontroller's memory to record all current online loads, i.e., the online load status list. The processing mechanism is that the microcontroller updates this list according to the direction of power change. If the power increases, the newly identified load information (type and fingerprint) is added to the list as a new entry; if the power decreases, the entry that best matches its characteristics is searched and removed from the list. The composite safety baseline is generated by arithmetically summing the steady-state root-mean-square current characteristic values ​​of all loads in the online load status list, and combining this with the expected surge current characteristics of inductive loads to calculate the composite current upper limit and waveform characteristics. Specifically, the generation method is as follows: first, the steady-state root-mean-square current of all loads in the list is arithmetically summed. The composite steady-state root-mean-square current is obtained by summing the results. Secondly, to predict potential combined surges, the composite peak surge current is calculated by multiplying the composite steady-state root-mean-square current by 1.414, and then adding the maximum additional surge current that may occur among all online inductive loads in the list—that is, the maximum difference between its surge peak and its own steady-state peak. This ensures that when an inductive load starts up, it will not be mistakenly judged as an overall overload due to a sudden surge in total current. For waveform characteristics such as harmonics and crest factors, the worst-case principle can be adopted. For example, the composite crest factor can be set to the maximum crest factor of all loads in the list. The output is a composite safety baseline representing the current overall operating state of multiple loads. This composite safety baseline is then used to replace the dynamic safety baseline of a single load, serving as a new basis for real-time monitoring and anomaly judgment, thereby achieving adaptive and accurate monitoring of complex multi-load scenarios.

[0045] In this embodiment of the application, the early warning capability for equipment failure can also be added. The intelligent triggering logic further includes: when the offset between the real-time harmonic component characteristics and the corresponding harmonic component characteristics in the dynamic safety baseline exceeds a preset threshold, it is determined that the equipment is abnormal.

[0046] Specifically, the microcontroller takes the real-time calculated current harmonic characteristic vector and the reference harmonic vector stored in the dynamic safety baseline (single or composite baseline) as inputs. It periodically calculates the difference between these two vectors, employing an effective vector distance or similarity metric, such as calculating the square of the Euclidean distance between two normalized harmonic vectors as the harmonic offset. Then, the calculated harmonic offset is compared with a preset harmonic offset threshold, which is statistically derived from extensive experimental data and characterizes the critical point of harmonic changes from normal aging to early failure in the equipment. For example, in a switching power supply, the capacitance decay of its internal electrolytic capacitors can lead to significant and predictable changes in its third and fifth harmonic components. When these predictable changes accumulate to a certain extent, even if the total power consumption does not increase significantly, it indicates that the device is about to fail. When the harmonic offset exceeds the harmonic offset threshold, the system determines that the device is in an abnormal fault state and triggers a warning signal that is different from an overload alarm. For example, it may use an LED to flash in a specific color or push a notification to the user's mobile application that "the device health has decreased and it is recommended to check it," thereby providing a warning before a catastrophic failure occurs.

[0047] Understandably, in order to achieve system-level coordinated protection for the entire circuit branch, the method also includes a coordinated protection step based on a wireless mesh network: a wireless mesh network covering a specific physical space is automatically constructed by a low-power wireless communication module integrated in the power line monitoring point; each monitoring point in the network periodically broadcasts data packets containing its own load status information to the network; a central node receives and aggregates load status information from multiple monitoring points belonging to the same circuit group as preset by the user to calculate the total load of the circuit group; the calculated total load is compared with the rated threshold of the circuit corresponding to the circuit group, and when the total load approaches or exceeds the rated threshold of the circuit, an early warning or protection command is executed.

[0048] Specifically, at the hardware level, each monitoring unit, in addition to the microcontroller and current acquisition device described in the above embodiments, also integrates a low-power wireless communication module, such as a wireless system-on-a-chip that supports the Bluetooth Low Energy mesh network protocol. When multiple such monitoring units are powered on in the same physical space, they will automatically execute a networking process. A unit designated as the central node, a dedicated gateway device, or a user's smartphone application, acts as the network configurator, configuring all nodes into the same mesh network, allocating network keys and application keys, and forming a robust wireless network with self-healing capabilities, no central authority, and multi-path transmission.

[0049] Each monitoring point (node) in the network acts as a data publisher, periodically performing a broadcast operation, for example, every 5 seconds. The input to this process is the load status information calculated locally by the node. The data packet content of the load status information includes at least: a unique identifier of the monitoring point device, the current load power, the load type, and the current safety margin percentage. Specifically, the binary structure of a data packet is defined as: a 4-byte unique identifier; a 2-byte unsigned integer representing the current load power in watts; a 1-byte enumeration value representing the identified load type, such as 0x01 representing resistive and 0x02 representing inductive; and a 1-byte signed integer representing the current safety margin percentage. This safety margin percentage is calculated by dividing the difference between the local baseline RMS current and the real-time RMS current by the baseline RMS current and then multiplying by 100, directly reflecting the local load pressure. This highly condensed data packet is broadcast by the node to a preset group address via a wireless module. All nodes in the network, including the central node, subscribe to this address.

[0050] The central node continuously receives broadcast data packets from all nodes in the network. The input consists of these data packet streams and a circuit group configuration information pre-defined by the user through a supporting application. This configuration information is a mapping table that assigns the unique identifiers of multiple monitoring points physically connected to the same air switch to the same circuit group identifier. The central node maintains a real-time status table in its memory, using the unique identifier as an index, to store the latest load status of each node. The central node periodically, or each time it receives a data packet update, traverses the circuit group configuration table. For each circuit group, it sums the current load power values ​​of all nodes in the group to obtain the total load power of that circuit branch.

[0051] Finally, the central node performs circuit-level comparisons and decisions. The inputs are the calculated total load power of the circuit branch and the preset rated power threshold for that circuit. For example, for a 16-ampere circuit breaker, its rated power is 3520 watts. The central node compares the total load power of the circuit branch with the rated power threshold and sets an 85% warning power threshold, i.e., 2992 watts. If the total load power of the circuit branch exceeds the warning power threshold but is lower than the rated power threshold, the central node broadcasts a warning command to all nodes within that circuit group, and the node's LED flashes yellow. If the total load power of a circuit branch exceeds the circuit's rated power threshold, the central node executes a protection command. It can broadcast a general disconnect command or execute a more intelligent strategy based on the load type information reported by each node. For example, it can prioritize sending targeted disconnect commands to nodes connected to resistive loads to unload non-critical loads and attempt to reduce the total load to a safe range without interrupting the operation of critical inductive loads such as refrigerators. This achieves overload prevention and collaborative management from single-point overload protection to the entire branch level, effectively avoiding branch tripping caused by the simultaneous operation of multiple high-power devices.

[0052] It is understandable that cloud intelligence can also be introduced to achieve the system's self-evolution capability. The method also includes a cloud enhancement step: with user authorization, the locally generated load digital fingerprint and load change event data are uploaded to the cloud server in an anonymized manner; the cloud server uses big data analysis and machine learning algorithms to train on the massive uploaded data to optimize and expand the general appliance fingerprint database; and the updated general appliance fingerprint database is pushed to the monitoring points in the network through online firmware upgrades.

[0053] Specifically, the central node, acting as a gateway between the local network and the cloud, after obtaining explicit authorization from the user, rigorously anonymizes newly learned payload fingerprints that cannot be identified by existing fingerprint databases. All information related to user identity and geographic location is stripped away, retaining only the pure technical data of the payload fingerprints. This data is then uploaded to a dedicated data lake on the cloud server via an encrypted hypertext transfer protocol. A machine learning pipeline is deployed in the background of the cloud server, which periodically processes massive amounts of anonymized fingerprint data collected from millions of devices worldwide. First, unsupervised clustering algorithms, such as density-based spatial clustering, are used to automatically discover dense clusters of device fingerprints in a multidimensional feature space. These clusters represent newly emerging appliance types on the market. Then, data scientists manually label these new device fingerprint clusters and use the entire labeled dataset to retrain a more powerful classification model, such as an efficient gradient boosting decision tree classification model or a lightweight neural network, which can more accurately map a fingerprint to a device type. After training, a general-purpose appliance fingerprint database containing more device models and with higher recognition accuracy is deployed to a cloud distribution network. The general-purpose appliance fingerprint database may be a model file or an updated parameter set. Finally, the central node will periodically query the cloud server for available fingerprint database updates. Once a new version is found, it can automatically download the update package and securely broadcast the update package in the form of small data blocks to all monitoring nodes in the local network through the firmware distribution mechanism provided by the Bluetooth Low Energy mesh network. Each node completes the update seamlessly in the background.

[0054] This application provides a real-time current monitoring system for overload protection power lines that implements the above-described method, comprising: a processor and a memory electrically connected to the processor; wherein, the memory stores computer-executable instructions, which, when executed by the processor, cause the system to perform the above-described method.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0056] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or 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, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0057] 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.

[0058] 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 functional unit.

[0059] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. 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 storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0060] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for real-time current monitoring of overload protection power lines, characterized in that, include: The current waveform data of the load on the power line is collected in real time using a current acquisition device at a preset sampling frequency. After a new load is detected and stabilized, a learning cycle is triggered. During the learning cycle, multi-dimensional feature parameters of the current waveform data are extracted. The multi-dimensional feature parameters include: peak surge current, surge duration, steady-state root mean square current, current crest factor, and harmonic component characteristics. The triggering condition for the learning cycle is: after a significant step increase in current from zero or a stable base value is detected, the root mean square value of the current fluctuates less than a preset stability threshold within a preset stable observation period. The multidimensional feature parameters are combined to generate a load digital fingerprint that characterizes the normal operation mode of the load, and the load digital fingerprint is stored in non-volatile memory as a dynamic security baseline. When multiple loads are connected to the power line, the system monitors the rate of change of total power to identify power step changes and defines the power step change as a load change event. When the load change event occurs, the difference between the current waveform data before and after the event is extracted to generate a difference digital fingerprint. The difference digital fingerprint is matched with a preset general electrical fingerprint database to identify the load type that has changed. Based on the identification result, an online load status list is updated, which records all currently online load types. Based on the online load status list, a composite safety baseline for overall risk assessment is dynamically generated. The composite safety baseline is generated by arithmetically summing the steady-state root mean square current characteristic values ​​of all loads in the online load status list and combining them with the expected surge current characteristics of the inductive load to calculate the composite current upper limit and waveform characteristics. After the learning cycle ends, real-time current waveform data is continuously collected, and its corresponding real-time multidimensional feature parameters are calculated. The real-time multidimensional feature parameters are compared with the stored dynamic security baseline, and an anomaly is determined when at least one of the following preset logic conditions is met: When the real-time steady-state root mean square current continues to exceed the preset percentage of the corresponding steady-state root mean square current in the dynamic safety baseline for a preset duration, it is determined to be a slow overload anomaly. When the real-time peak surge current exceeds the preset percentage of the corresponding peak surge current in the dynamic safety baseline, it is determined to be an instantaneous startup anomaly. When the offset between the real-time harmonic component characteristics and the corresponding harmonic component characteristics in the dynamic safety baseline exceeds a preset threshold, the equipment is determined to be faulty. A wireless mesh network covering a specific physical space is constructed by integrating a low-power wireless communication module into the power line monitoring point. Each monitoring point in the wireless mesh network periodically broadcasts data packets containing its own load status information to the network. A central node receives and aggregates the load status information from multiple monitoring points within the same circuit group to calculate the total load of the circuit group. The calculated total load is compared with the rated threshold of the circuit corresponding to the circuit group, and when the total load exceeds the rated threshold, an early warning or protection command is executed.

2. The method according to claim 1, characterized in that, The general electrical fingerprint database contains feature models for at least three load types: resistive load model, inductive load model, and switching power supply load model.

3. The method according to claim 1, characterized in that, The data packet containing the load status information includes at least the following: monitoring point device ID, current load power, load type, and current safety margin percentage.

4. The method according to claim 1, characterized in that, The method further includes: The load digital fingerprint and load change event data are uploaded to the cloud server in an anonymized manner; The cloud server uses big data analysis and machine learning algorithms to train the uploaded data in order to optimize and expand the general electrical appliance fingerprint database. The updated GE fingerprint database is pushed to monitoring points in the network via online firmware upgrades.

5. The method according to claim 1, characterized in that, The triggering condition for the learning cycle is: after the current is detected to rise from zero or a stable base value, the root mean square value of the current fluctuates less than a preset stable threshold during a preset stable observation period.

6. A real-time current monitoring system for overload protection power lines, characterized in that, include: processor; as well as The memory electrically connected to the processor; The memory stores computer-executable instructions, which, when executed by the processor, cause the system to perform the method as described in any one of claims 1 to 5.

Images