A campus lithium battery charging load identification system and method based on differential equivalent impedance spectrum and multi-physical domain intelligent sensing

Abstract

The application relates to the technical field of low-voltage power distribution safety, electrical load identification and campus power safety management, and discloses a campus lithium battery charging load identification system and method based on differential equivalent impedance spectrum and multi-physical domain intelligent sensing. The system obtains the differential equivalent impedance spectrum through differential calculation of the load access response and the node baseline response, identifies line abnormalities such as zero-line sharing and illegal cross connection through a sweep signal with node coding and a multi-node coupling spectrum matrix, and performs reliability gating determination by fusing electrical, thermal, acoustic and topological characteristics, and is suitable for power safety management of campus dormitories, teaching buildings, laboratories and centralized charging areas.

Description

Technical Field

[0001] This invention relates to the fields of low-voltage power distribution safety, electrical load identification, and campus power safety management, specifically to a lithium battery charging load identification system and method applicable to campus dormitories, teaching buildings, laboratories, electric bicycle charging areas, and centralized charging areas. Background Technology

[0002] The use of lithium battery devices on campus is increasing. Dormitories are filled with mobile phones, tablets, laptops, and power banks; some students even bring electric bicycle lithium batteries and portable power banks indoors to charge. Laboratories and teaching buildings also see instances of temporary charging and concentrated charging of mobile devices. While these uses themselves may not necessarily be illegal, when combined with unauthorized wiring, substandard chargers, modified lithium batteries, and high-power resistive electrical appliances, they pose a significant risk of fire and electric shock to low-voltage power distribution circuits.

[0003] In campus management, common risky loads can be broadly categorized into several types. The first type includes high-power resistive loads such as electric kettles, electric hot pots, induction cookers, and electric heaters. These devices have high power consumption and rapid temperature rise, making them easily detectable using power threshold methods. The second type includes substandard chargers, modified charging heads, and switching power supplies without proper protection circuits. While their power consumption may not be high, they can exhibit issues such as large output ripple, lack of input filtering, and overheating. The third type consists of abnormal loads where lithium batteries and electrical equipment are integrated, such as modified electric bicycle lithium batteries, substandard portable energy storage devices, and high-power lithium battery-powered lighting fixtures. The electrical behavior of these devices is complex and difficult to diagnose solely based on power or harmonics. The fourth type is not the load itself, but rather abnormal wiring connections, such as shared neutral wires, cross-circuit power borrowing, unauthorized wiring across floors, suspended ground wires, and poor contact in socket contacts. These problems are often more concealed than those involving individual loads.

[0004] Existing campus power management technologies mainly include manual inspection, power threshold-based power outages, waveform feature analysis, power fingerprinting, machine learning recognition, and tiered current pulse recognition. Manual inspections have discontinuous coverage and rely on the experience of management personnel; the power threshold method is simple and reliable, but can only effectively detect high-power equipment; waveform and power fingerprinting methods can identify some load types, but are prone to confusion when multiple loads are superimposed; machine learning methods require a large number of samples, and the model is prone to inaccuracy after changes in the field environment; tiered current pulses can obtain a small number of response points, but the amount of information is still limited.

[0005] Existing methods for identifying complex impedance can estimate the equivalent impedance of a load through voltage and current responses, which is a further improvement over simple power thresholding. However, in campus buildings, socket circuits vary in length and wiring methods, and coupling may exist between multiple sockets in the same dormitory. Background harmonics in the power grid and adjacent loads can also affect the measurement results. If the impedance spectrum after connection is directly used as the load characteristics, the result is often not a "clean fingerprint" of the load itself, but rather a result of the superposition of the load, the circuit, and the environment. For review and implementation, the real problem to be solved here is: how to extract the characteristics of the load itself from the complex campus circuits as much as possible without affecting normal power use.

[0006] Furthermore, campus security management needs to know not only "what devices are connected," but also "whether the wiring has been modified." Even if a single-point smart socket can identify the load type, it is difficult to detect abnormal electrical coupling between adjacent sockets, let alone determine whether two floors or circuits that should not be connected have been illegally cross-connected. Existing temperature-protected sockets usually only shut off power when overheating, failing to consider temperature rise trends, impedance spectra, acoustic events, and node coupling relationships in a comprehensive assessment. Acoustic sensors are often used for simple abnormal sound alarms and are rarely used in conjunction with relay actions from lithium battery charger battery management systems, PWM whistling, socket sparking sounds, and impedance spectrum results for load identification.

[0007] Therefore, it is necessary to propose a technical solution more suitable for campus socket networks. Ideally, this solution should achieve the following: the active detection signal should be sufficiently small to not affect normal power consumption; the impedance spectrum should be able to subtract the node baseline to reduce parasitic effects on the line; adjacent nodes should be able to identify the coupling response of the same sweep frequency signal; the system should be able to judge electrical, thermal, acoustic, and topology information together; and the system should still provide a conservative and reliable safety judgment when a sensor malfunctions or there is significant ambient noise. Summary of the Invention

[0008] The purpose of this invention is to provide a campus lithium battery charging load identification system and method based on differential equivalent impedance spectroscopy and multi-physical domain intelligent sensing. This system obtains frequency domain characteristics closer to the load itself through differential equivalent impedance spectroscopy scanning, identifies the electrical coupling relationship between adjacent nodes through coded frequency sweeping collaborative detection, improves the ability to distinguish similar loads through electrical, thermal, and acoustic multi-physical domain sensing, and reduces the risk of misjudgment caused by sensor anomalies and on-site interference through reliability-gated fusion judgment.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a campus lithium battery charging load identification system. The system includes a three-level identification layer, a differential equivalent impedance spectroscopy scanning module, an coded frequency sweep collaborative detection network, a multi-physical domain intelligent sensing module, a reliability gating fusion judgment unit, and an intelligent management and control platform. The three-level identification layer is responsible for processing detection tasks separately according to scenarios and risk levels; the differential equivalent impedance spectroscopy scanning module is responsible for extracting the frequency domain impedance characteristics of the load; the coded frequency sweep collaborative detection network is responsible for connecting multiple detection nodes into a mutually verifiable sensing network; the multi-physical domain intelligent sensing module is responsible for supplementing temperature rise, acoustic, and conventional electrical characteristics; the reliability gating fusion judgment unit is responsible for comprehensive judgment; and the intelligent management and control platform is responsible for alarms, power outages, recording, and subsequent operation and maintenance.

[0010] Furthermore, the three-level identification layer includes a first-level identification unit, a second-level identification unit, and a third-level identification unit. The first-level identification unit is located at the front end of ordinary sockets, primarily processing basic parameters such as power and current, and can directly cut off power when a clearly high-power illegal load is detected. The second-level identification unit is located in key areas such as centralized charging areas, electric bicycle charging sheds, and laboratory lithium battery charging areas, primarily performing differential equivalent impedance spectroscopy scanning. The third-level identification unit is located at key charging ports and sockets requiring close monitoring; in addition to performing impedance spectroscopy detection, it is also responsible for initiating coded frequency sweep collaborative detection to determine whether there is abnormal coupling between adjacent nodes.

[0011] Furthermore, the differential equivalent impedance spectrum scanning module does not simply measure the impedance value after connection. Instead, it first establishes a node baseline, then subtracts the node baseline response from the response after load connection to obtain the differential response. This is done because factors such as line length, terminal block conditions, socket aging, and adjacent loads in campus buildings can all cause offsets in the impedance spectrum. Without subtracting the baseline, the model might mistakenly identify "line differences" as "load differences." The node baseline can be derived from the real-time baseline when the current node is unloaded, from a built-in reference branch, or from the calibration baseline of adjacent idle nodes, with cable length and temperature compensation applied.

[0012] Furthermore, the differential equivalent impedance spectrum scanning module injects a low-amplitude differential sweep frequency signal into the live and neutral wire circuits via isolated coupling. The sweep frequency range is preferably 100Hz to 100kHz, with the sweep amplitude not exceeding 2% of the current supply voltage's effective value and the peak-to-peak value not exceeding 5V. This amplitude is sufficient for the sampling circuit to extract the response without causing power fluctuations for the user or substantially affecting the normal operation of the charger. When user plugging / unplugging, sudden large current changes, or abnormal fluctuations in mains voltage are detected, the module pauses the sweep frequency and resumes detection once the circuit stabilizes.

[0013] Furthermore, the coded sweep frequency collaborative detection network assigns a unique node code to each level-three identification unit. The initiating node superimposes its own node code onto the sweep frequency signal, while adjacent nodes acquire the voltage and current waveforms of their respective circuits within a synchronization time window, and then perform matched filtering using the corresponding coding template. In this way, adjacent nodes can not only know "whether there is a coupling signal," but also "which node this coupling signal comes from." This design avoids mutual confusion when multiple nodes detect simultaneously and also provides a data foundation for the subsequent construction of the coupling spectrum matrix.

[0014] Furthermore, the coded frequency sweep collaborative detection network constructs a multi-node coupling spectrum matrix based on the coupling responses between nodes. Each item in the matrix reflects the attenuation and phase change when a detection signal is propagated from one node to another. If the coupling response of adjacent nodes within the same circuit attenuates normally with increasing distance, it is determined to be normal line coupling; if continuous high coupling occurs between different floors or different power distribution circuits, and the connection is not within the allowable range of the campus power distribution topology, it is determined to be shared neutral wire, unauthorized bridging, abnormal insulation coupling, or other line topology abnormalities.

[0015] Furthermore, the multi-physics domain intelligent sensing module includes an electrical feature acquisition submodule, a thermal feature acquisition submodule, an acoustic feature acquisition submodule, and a sensor status self-testing submodule. The electrical feature acquisition submodule acquires conventional data such as voltage, current, active power, power factor, and total harmonic distortion (THD). The thermal feature acquisition submodule acquires temperature changes using a negative temperature coefficient thermistor attached near the copper contacts or spring contacts of the socket. The acoustic feature acquisition submodule acquires internal vibrations and acoustic events of the socket using a piezoelectric ceramic plate. The sensor status self-testing submodule determines whether each channel is disconnected, saturated, has excessive noise, or exhibits abnormal response.

[0016] Furthermore, the reliability-gated fusion decision unit evaluates the reliability of each physical domain feature before fusion. When the acoustic noise in the field is high, the weight of acoustic features decreases; when the thermistor is broken or poorly attached, the weight of thermal features decreases; when the sweep frequency signal-to-noise ratio is insufficient, the weight of impedance spectrum features decreases. Only features that pass the reliability gating are included in the fusion decision model. This model can be a lightweight Transformer encoder or an equivalent classification model capable of handling multi-domain features.

[0017] Furthermore, the reliability gating fusion judgment unit also incorporates physical rule fallback logic. When the model classification confidence is low, if the system has already detected high temperature rise, arcing acoustic events, or abnormal cross-loop coupling edges, it will not directly output "safe," but will first output a risk warning. This logic aligns with the safety requirement in campus electricity management: "It is better to prompt investigation than to overlook high-risk evidence."

[0018] This invention also provides a method for identifying lithium battery charging loads on campus. The method includes steps such as initialization, primary power identification, differential equivalent impedance spectrum scanning, coded frequency sweep collaborative detection, coupled spectrum topology determination, multi-physical domain feature acquisition, reliability gating fusion determination, and intelligent management. These steps are not simply connected in series, but are triggered in layers according to risk level: obviously high-power loads are directly handled by the primary identification unit; loads with low power but complex characteristics enter the impedance spectrum and multi-physical domain fusion determination; scenarios requiring line relationship determination are initiated by the coded frequency sweep collaborative detection unit.

[0019] Beneficial effects Compared with the prior art, the present invention has at least the following beneficial effects: First, this invention further processes ordinary complex impedance identification into differential equivalent impedance spectrum identification. By subtracting the load access response from the node baseline response, the impact of line parasitism, node differences, power grid background, and coupling of adjacent loads on the identification results is reduced. For usage scenarios of the same model charger in different dormitories, different floors, and different cable lengths, this processing can improve feature consistency.

[0020] Second, this invention employs frequency sweep collaborative detection with node coding. Adjacent nodes identify the coded frequency sweep signal of the initiating node through matched filtering, forming traceable coupling relationship data among multiple nodes. This data is not only used to assist load identification but also to detect line topology anomalies such as shared neutral wires and unauthorized cross-circuit connections.

[0021] Third, this invention places five types of information—electrical, differential equivalent impedance spectrum, thermal, acoustic, and topological—under the same judgment framework. Resistive high-power loads typically exhibit high temperature rise and low impedance change, while standard lithium battery chargers usually have input filter resonance characteristics and low temperature rise. Poor contact is often accompanied by temperature steps and short-duration acoustic pulses. The mechanisms of different physical domains are different, which can compensate for the problem of overlapping single electrical characteristics.

[0022] Fourth, this invention employs reliability-gated fusion. When a sensor fails, noise levels are too high, or the signal saturates, the system will not equate the abnormal data from that channel with valid evidence, but will instead reduce its weight or switch to a conservative strategy. This improves the long-term stability of the system.

[0023] Fifth, the output of this invention includes not only load type, but also line topology anomaly risk, thermal anomaly risk, and acoustic anomaly risk. The management platform can then generate tiered alarms, remote power outages, work order investigations, and risk heatmaps, better meeting the actual management needs of campus scenarios. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall system architecture provided in an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the differential equivalent impedance spectrum scanning module provided in an embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of the coded frequency sweeping collaborative detection network provided in an embodiment of the present invention.

[0027] Figure 4 This is a schematic diagram of multi-node coupling spectrum matrix and topological anomaly determination provided in an embodiment of the present invention.

[0028] Figure 5 This is a schematic diagram of the installation structure of the multi-physics domain intelligent sensing module provided in an embodiment of the present invention.

[0029] Figure 6 This is a schematic diagram of a reliability gating fusion determination unit provided in an embodiment of the present invention.

[0030] Figure 7 This is a schematic diagram of the differential equivalent impedance spectrum of different types of loads provided in the embodiments of the present invention.

[0031] Figure 8 This is a schematic diagram of the identification method provided in an embodiment of the present invention. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The following embodiments are used to illustrate the implementation of the present invention and are not intended to limit the scope of protection. Those skilled in the art can make equivalent substitutions for specific device models, communication protocols, thresholds, sampling frequencies, encoding sequence lengths, and model structures without departing from the concept of the present invention.

[0033] Example 1: The connection between the main system and its modules Reference Appendix Figure 1 The system in this embodiment includes a three-level identification layer 100, a differential equivalent impedance spectrum scanning module 200, an coded frequency sweeping collaborative detection network 300, a multi-physical domain intelligent sensing module 400, a reliability gating fusion judgment unit 500, and an intelligent management and control platform 600.

[0034] The input to the third-level identification layer 100 comes from electrical sampling data and on / off status of each socket or charging port. The first-level identification unit 110 mainly deals with ordinary sockets. Its processing logic is very straightforward: first, check for obvious overpower or overcurrent. If it exceeds the threshold and persists for a period of time, the risk is considered sufficiently clear, and without waiting for a complex model to make a judgment, the power is directly cut off and an alarm is reported. The output of the first-level identification unit 110 includes basic electrical characteristics, on / off control results, and first-level alarm information.

[0035] The secondary identification unit 120 targets key areas such as centralized charging zones. Many substandard chargers and modified charging devices have low power outputs, and the primary identification unit 110 does not directly process them. Therefore, the secondary identification unit 120 receives the load status after the primary identification unit 110 has approved it and activates the differential equivalent impedance spectrum scanning module 200. Its output is not a simple power value, but a set of frequency domain characteristics, including low-frequency impedance, resonant peaks, phase changes, and impedance amplitude variations.

[0036] The Level 3 Identification Unit 130 is designed for key monitored ports. It can perform the functions of the Level 2 Identification Unit 120 and also act as the initiating node in the coded frequency sweep collaborative detection network 300. When it is necessary to determine whether there are any line topology anomalies, the Level 3 Identification Unit 130 broadcasts a detection start message to neighboring nodes and injects a frequency sweep signal with node coding. After neighboring nodes return coupling responses, the Level 3 Identification Unit 130 or the intelligent management platform 600 constructs a coupling spectrum matrix and outputs abnormal coupling edges and topology risks.

[0037] The multi-physics domain intelligent sensing module 400 and the three-level recognition layer 100 work in parallel. Instead of simply adding a temperature value at the end of the model, it continuously collects electrical, thermal, and acoustic data throughout the entire process after the load is connected, and assigns a unique timestamp to each data point. This allows the system to determine whether a temperature step occurs after the load is connected, whether an acoustic pulse corresponds to socket arcing, and whether a BMS relay acoustic event matches the charger characteristics in the impedance spectrum.

[0038] The reliability gating fusion judgment unit 500 receives the output of the aforementioned modules. It first checks whether the data is reliable, then assigns weights according to reliability, and finally outputs the load type, violation probability, and line topology anomaly level. After receiving the above results, the intelligent management and control platform 600 executes graded alarms, remote power outages, administrator notifications, work order generation, baseline spectrum updates, and risk area display.

[0039] The key feature of this main system is that the modules are not stacked in isolation, but rather layered according to the risks of campus electricity use. Level 1 identification solves the problem of obviously high power; impedance spectroscopy solves the problem of low power but complex circuit characteristics; coded frequency sweep solves the problem of line connection problems that cannot be seen by single-point detection; thermoacoustic sensing fills the judgment gap caused by the overlap of electrical characteristics; and reliability gating handles the noise and sensor anomalies that are unavoidable in long-term field operation.

[0040] Example 2: Input, Processing, and Output of the Differential Equivalent Impedance Spectrum Scanning Module Reference Appendix Figure 2The differential equivalent impedance spectrum scanning module 200 includes a sweep frequency signal generator, an isolation coupling unit, a current limiting protection unit, a dual-channel synchronous sampling unit, a baseline switching unit, and an impedance spectrum processing unit. Its inputs include three types: the first type is the current electrical state of the node, such as the effective value of voltage, the effective value of current, and the on / off state of relays; the second type is the node baseline spectrum, such as the no-load baseline, the reference branch baseline, or the calibration baseline of adjacent nodes; and the third type is the sweep frequency response after the load is connected.

[0041] In campus sockets, the impedance spectrum varies when the same load is plugged into different locations. This is not due to changes in the load itself, but rather to variations in cable length, terminal blocks, nearby loads, and power frequency background. Therefore, this embodiment does not directly use the voltage-to-current ratio after connection as a classification feature. Instead, it first obtains the node baseline and then performs differential calculations. The technical problem this step solves is to minimize the "offset caused by the node itself" from the "response of the load itself."

[0042] At a swept frequency, the impedance spectrum processing unit first obtains the complex voltage and complex current responses under the load-connected state, and then obtains the complex voltage and complex current responses under the node baseline state. To extract the changes relative to the baseline after load connection, the following formula is used: ; in, Indicates angular frequency as Differential equivalent impedance at time, Represents angular frequency. This represents the difference in complex voltage response. This represents the difference in complex current response. The purpose of this formula is to guide the system to focus on the response changes caused by load connection, rather than on the original background response of the node.

[0043] To obtain the two difference components mentioned above, the system further employs the following calculation: ; ; in, This represents the complex voltage response under load conditions. This represents the complex voltage response under the node baseline state. This represents the complex current response under load conditions. These two formulas represent the complex current response under the node baseline state. By using these two formulas, the system can extract the changes before and after the load is connected, making it less likely for the subsequent classification model to mistake cable aging or socket structural differences for load characteristics.

[0044] The source of the node baseline can be selected according to the actual scenario. If the socket is idle, directly collecting the no-load baseline is the most accurate; if the socket is constantly connected to equipment, the built-in reference branch can be switched to obtain the online baseline; if the current node cannot be switched, the baseline can also be taken from an adjacent idle node on the same floor and in the same circuit, and then compensation can be performed. For baseline compensation of adjacent nodes, the technical problem faced by the system is that the cable lengths and temperatures of different nodes are different, so the baselines cannot be used directly. Therefore, the following compensation methods can be adopted: ; in, Indicates conversion to the th The equivalent baseline of the node. Indicates the first The measured baseline of the node, Indicates by the first Node cable length and the Node cable length Determined length compensation coefficient, Indicates by the first Node ambient temperature and the Node ambient temperature A defined temperature compensation coefficient. This formula allows the baselines of adjacent nodes to be used as backup baselines, preventing detection failures due to a lack of available windows for the current node.

[0045] After calculating the differential equivalent impedance at each frequency point, the system needs to transform the entire spectrum into features that are easy to classify. For a standard charger with an input filter network, a resonance peak often appears in a specific frequency band; for a purely resistive load, the impedance amplitude is usually relatively flat, and the phase angle is close to zero. To describe the location of the resonance peak, the following formula can be used: ; in, Indicates the characteristic resonant peak frequency. Represents pi (π). This represents the equivalent inductance of the load input filter network. This represents the equivalent capacitance of the load input filter network. This formula does not require the system to know the actual component parameters, but rather illustrates the correlation between the resonance peak and the input filter structure. The system locates the resonance peak through measured spectrum analysis. This allows you to distinguish between a legitimate lithium battery charger and a substandard charger that lacks a proper filtering structure.

[0046] To describe whether a resonance peak is sharp, a quality factor is used: ; in, Indicates the quality factor. Indicating the resonance peak bandwidth. A higher value indicates a sharper resonance peak, which usually corresponds to a more defined input filter network. A low or no obvious resonance peak may correspond to a poor-quality charger, an abnormal switching power supply, or a resistive load.

[0047] To describe whether the impedance amplitude is flat across the entire frequency band, the coefficient of variation is used: ; in, Indicates the coefficient of variation of impedance amplitude across the entire frequency band. This represents the standard deviation of the differential equivalent impedance magnitude. This represents the average value of the differential equivalent impedance. Purely resistive loads typically exhibit small amplitude variations over a wide frequency range. Low; loads containing reactive components and filtering structures, It usually increases.

[0048] The output of this module includes the differential equivalent impedance spectrum curve, the average impedance amplitude in the low-frequency band, the characteristic resonant peak frequency, the quality factor, the high-frequency phase angle slope, the impedance amplitude variation coefficient, and the sweep frequency signal-to-noise ratio. If user plugging or unplugging, relay operation, sudden large current, or voltage fluctuation occurs during the detection, the module outputs a "sweep invalid" flag and excludes this data from the model input. The detection will be repeated after the circuit stabilizes.

[0049] Example 3: Coordinated Detection by Encoded Frequency Sweep and Coupled Spectral Matrix Reference Appendix Figure 3 The coded sweep frequency collaborative detection network 300 addresses the problem of single-point detection not showing the line connection relationship. Each three-level identification unit has a unique node address and node code. The node code can be a pseudo-random binary sequence or an orthogonal spreading code. The initiating node superimposes its own node code onto the sweep frequency signal, and adjacent nodes identify this code through matched filtering, thus determining which initiating node the coupled signal originated from.

[0050] In a campus setting, detection may occur simultaneously in multiple dormitories and on multiple floors. Without encoding, adjacent nodes only see disturbances at a certain frequency, making it difficult to determine their origin and easily mistaking external noise for coupled signals. Therefore, an identifiable detection signal is needed. The initiating node is at the... The coded sweep signal injected at each sweep frequency point can be represented as: ; in, Indicates the first The initiating node is in the... Encoded sweep frequency signal at each frequency point Indicates time, Indicates the first Injection amplitude at each frequency point, Indicates the first The encoded sequence of each node, Indicates the first Each sweep frequency point Indicates the first The initial phase of each node. This formula binds frequency characteristics to node identity, enabling neighboring nodes to identify the signal source from the sampled waveform.

[0051] Adjacent nodes acquire waveforms within the synchronous monitoring window. Then, matched filtering is performed using the initiating node's encoding template. To reduce the impact of random noise, the correlation quantity can be calculated: ; in, Indicates the first The nth adjacent node pair The initiating node is in the... Matched filtering results at each frequency point Indicates the start time of the synchronous monitoring window. Indicates the length of the synchronous monitoring window. Indicates the first The waveforms acquired by each node. If Exceeding the preset threshold indicates that the first The node did indeed receive the first... The coupling signal of each node.

[0052] After each adjacent node transmits back the amplitude, phase, and arrival time of the coupled signal, the system constructs a coupling spectrum matrix. The elements in the coupling spectrum matrix can take the following form: ; in, Indicates the first node to the Node at frequency point The coupling spectrum value at that location, Indicates the first The node received the first The complex response after node-encoded frequency sweep signal, Indicates the first The node itself references the response. This represents the logarithm to base 10. The coupling spectrum value is expressed in decibels (dB), facilitating comparisons of the isolation levels between different nodes.

[0053] To avoid misjudgments caused by occasional anomalies at a single frequency point, the system also calculates the average anomaly level over a specified frequency band: ; in, Indicates the first node to the Node at Average coupling at each frequency point This indicates the number of frequency points included in the statistics. If It consistently exceeds the reference range for normal nodes of the same type, and the first Node and the If a node should not have a direct coupling relationship in the original power distribution topology of the campus, the system will mark it as an abnormal coupling edge.

[0054] When determining topology anomalies, the system considers not only numerical values ​​but also the campus power distribution diagram and node locations. Under normal circumstances, adjacent nodes within the same room or circuit may have limited coupling, which gradually decreases with increasing distance. Sustained high coupling between different floors or different power distribution circuits is abnormal. If abnormal coupling is accompanied by a temperature step or arcing acoustic event, the line risk level is increased. This reduces false alarms caused by environmental electromagnetic interference.

[0055] The output of this embodiment includes a coupling spectrum matrix, a list of abnormal coupling edges, suspected anomaly types, related node numbers, risk levels, and suggested troubleshooting paths. The intelligent management platform can display abnormal coupling edges on the floor plan, allowing managers to know which sockets or distribution boxes should be checked first.

[0056] Example 4: Multi-Physical Domain Intelligent Sensing Module Reference Appendix Figure 5 The multi-physics domain intelligent sensing module 400 does not replace electrical detection, but rather supplements those risks that are difficult to distinguish using electrical characteristics. Its inputs include routine electrical sampling, socket metal temperature, internal acoustic vibration signals of the socket, and the self-test status of each sensor. Its outputs include electrical feature vectors, thermal feature vectors, acoustic feature vectors, and sensor reliability indicators.

[0057] The electrical characteristic acquisition submodule 410 can be implemented using a power metering chip. It continuously acquires voltage, current, active power, power factor, and harmonic information. When a load is first connected, the electrical characteristics can quickly reflect startup surges, power level changes, and power factor variations; once the load is operating stably, the electrical characteristics can reflect the continuous power consumption status. Abnormal scenarios for this submodule include sampling chip communication failure, voltage sampling loss, and current transformer saturation. If these abnormalities occur, the system reduces electrical domain reliability and prompts for maintenance.

[0058] The thermal signature acquisition submodule 420 uses a negative temperature coefficient thermistor attached near the copper contacts or spring contacts of the socket. The problem that the thermal path needs to address is that some resistive loads or poor contact risks are not always particularly prominent in terms of power values, but the temperature rise trend is obvious. To separate the temperature rise from the slow changes in ambient temperature, the system calculates the temperature rise rate within a time window after the load is connected: ; in, Indicates the rate of temperature rise. This indicates the temperature at the second sampling time. This indicates the temperature at the first sampling time. Indicates the second sampling time. This indicates the first sampling time. If the temperature rise rate is significantly higher than the normal range for similar sockets, the system determines that there is a high heat load or a risk of overheating. If the temperature curve shows a short-duration step, but the power does not increase significantly at the same time, it is more likely due to poor contact or spring arcing.

[0059] The acoustic feature acquisition submodule 430 uses a piezoelectric ceramic plate to collect internal vibration and acoustic signals from the socket. The acoustic channel is primarily used to capture three types of events: first, the relay closing sound in a standard lithium battery charger or battery management system; second, the PWM whistling sound generated by some chargers during the constant current phase; and third, short-duration pulse sounds caused by poor contact or arcing. To describe these sounds in both time and frequency dimensions, the system employs a short-time Fourier transform: ; in, Indicates the first Each time position and frequency is The time-frequency spectrum results at that location Represents an acoustic sampling sequence. Represents the window function. Indicates the length of the window. This indicates the imaginary unit. The technical function of this formula is to describe short-duration pulses, continuous howling, and relay operating sounds separately, rather than just looking at a single total sound pressure level.

[0060] The sensor status self-test submodule 440 performs health checks on each channel. The thermal channel checks whether the NTC resistance is within a reasonable range, the acoustic channel checks whether the background noise is saturated, the electrical channel checks whether the sampling is synchronized with the power frequency zero crossing, and the impedance spectrum channel checks the synchronous demodulation residual of the swept frequency signal. The self-test output enters the reliability gating fusion judgment unit 500, which determines whether the corresponding feature can participate in the final judgment and the degree of participation.

[0061] Example 5: Reliability Gated Fusion Decision Reference Appendix Figure 6The reliability-gated fusion judgment unit 500 receives inputs including electrical characteristics, differential equivalent impedance spectrum characteristics, thermal characteristics, acoustic characteristics, coupling spectrum topological characteristics, and sensor self-test results. Its processing logic consists of three layers: the first layer calculates reliability, the second layer performs weighted fusion based on reliability, and the third layer performs classification and risk mitigation.

[0062] In real-world campus environments, sensors are not always in ideal condition. For example, the noise level in dormitories can cause the acoustic channel to detect many irrelevant vibrations; after years of use, thermistors in electrical outlets may become loose; and large fluctuations in the power grid at certain times can affect impedance spectrum sweeps. If the system does not first determine the reliability of the data, the model can easily be skewed by bad data. Therefore, this embodiment first calculates the reliability of each physical domain: ; in, Indicates the first Reliability of physical domain-like features It can represent any of the following: electrical domain, impedance spectrum domain, thermal domain, acoustic domain, or topological domain. Indicates the first Signal-to-noise ratio of class features Indicates the preset signal-to-noise ratio reference. This represents the sensor's self-test effectiveness coefficient. Represents the data integrity coefficient. This indicates a function that takes the smaller value. The effect of this formula is that physical domains with insufficient signal-to-noise ratio, failed self-tests, or missing data will not be given excessive weight.

[0063] After the reliability calculation is completed, the system normalizes the reliability into fusion weights: ; in, Indicates the first Fusion weights for physical domain-like features This represents the sum of the physical domain reliability scores of all participating channels. After normalization, the sum of the weights of all participating channels is 1. When a channel in the field becomes unreliable, its weight naturally decreases, allowing other reliable channels to take on more decision-making roles.

[0064] Subsequently, the reliability-gated fusion decision unit 500 constructs the fusion feature vector: ; in, Represents the fused feature vector. Represents the electrical domain weights. Represents the electrical feature vector. Represents the impedance spectral domain weights. This represents the eigenvector of the differential equivalent impedance spectrum. Indicates the heat domain weight. Represents the thermal eigenvector. Represents the acoustic domain weights. Represents the acoustic feature vector. Represents the topological domain weights. This represents the topological eigenvector of the coupled spectrum.

[0065] The fusion model can employ a lightweight Transformer encoder. Features from each physical domain are first mapped to token vectors, and then reliability is encoded and added to the corresponding tokens, enabling the model to know not only "what this feature is," but also "whether this feature is currently reliable." The model outputs load type probabilities and topological risk probabilities. For schools with insufficient training samples, the model can initially run using a rule-based classifier, and then be gradually updated using real-world campus data.

[0066] To prevent safety risks from being masked by model uncertainties, the system also includes a fallback logic based on physical rules. If the model classification confidence level is below a preset threshold, but the thermal channel shows a high temperature rise, the acoustic channel shows a short-term ignition pulse, or the topology channel shows an abnormal coupling edge across a loop, the system will not output "safe load," but rather "risk needs to be investigated." This output can serve as a warning, not necessarily an immediate power outage, thus balancing safety and normal power usage experience.

[0067] Example 6: False Alarms, Abnormal Scenarios, and Conservative Handling Strategies The power supply situation on campus is very complex, and false alarm handling cannot rely solely on raising the threshold. This embodiment categorizes common anomalies into four types for handling.

[0068] The first type is false alarms caused by normal load superposition. For example, when multiple mobile phones, laptops, and desk lamps are connected at the same time, the power may approach the first-level threshold, and harmonics may also be superimposed. The system first checks for high temperature rise and abnormal coupling edge; if there is no high temperature rise, no abnormal acoustic event, and the differential equivalent impedance spectrum shows the combined characteristics of multiple low-power switching power supplies, then the system can output a "compliant superposition of power consumption prompt" instead of directly cutting off the power.

[0069] The second category is false acoustic alarms caused by environmental noise. Knocking on a desk, opening and closing doors, and moving belongings in a dormitory can all be detected by the piezoelectric ceramic sensor. The system uses STFT to determine whether the acoustic event has the short-duration, high-frequency pulse characteristic of socket arcing, and combines this with electrical waveforms and temperature steps for further judgment. The risk level is only increased when an acoustic event occurs simultaneously with an electrical or thermal anomaly. If there is only an isolated acoustic event, it is recorded but the power is not immediately cut off.

[0070] The third type is disturbance during the frequency sweep process. User plugging and unplugging of devices, relay operation, and power grid voltage fluctuations can all affect the impedance spectrum. When the system detects these anomalies, it marks the frequency sweep data as invalid and does not proceed to the fusion judgment. If multiple frequency sweeps are invalid consecutively, the system indicates that the node environment is unstable, and the administrator can choose to manually troubleshoot or retest after a delay.

[0071] The fourth category is sensor failure. This can be caused by a broken thermistor wire, a detached piezoelectric ceramic plate, or a communication malfunction in the sampling chip. The sensor status self-test submodule will reduce the reliability of the corresponding physical domain and report a maintenance alarm. In this case, the system can still continue to operate using the remaining channels, but the output will be marked "Degradation Decision". If high-risk evidence still appears after degradation, such as an abnormal cross-loop coupling edge, the system will still execute a safety warning.

[0072] Through the above strategy, the false alarm handling of this invention does not simply relax the threshold, but rather clearly distinguishes the sources of anomalies: normal power consumption, environmental noise, disturbances during the detection process, and sensor failure are handled separately. This reduces unnecessary power outages and prevents genuine line risks from being overlooked.

[0073] Example 7: Data Security, Baseline Updates, and Device Implementation This embodiment illustrates the data management and equipment implementation methods when the system is deployed on a large scale on campus.

[0074] The data uploaded by the detection nodes primarily consists of feature data, rather than continuously uploading raw acoustic waveforms. The acoustic channel only uploads features such as event time, frequency band energy, duration, and event category, unless the administrator actively enables short-term waveform retention during troubleshooting. This reduces network load and avoids collecting environmental sound content unrelated to electrical safety. Device authentication and encrypted communication are used between the nodes and the platform to prevent forged nodes from uploading false statuses.

[0075] The baseline spectrum library is managed by node. Each node stores the factory baseline, the initial baseline after installation, and the updated baseline during operation. Baseline updates do not always overwrite old data; instead, they are only performed when the node is confirmed to be idle, the environment stable, and the sensors have performed normal self-tests. If an updated baseline deviates significantly from historical baselines, the system does not use it immediately but marks it as pending review. This prevents abnormal line conditions from being incorrectly written into the baseline library.

[0076] In terms of hardware, the first-level identification unit can be implemented using a power metering chip, a microcontroller, and a magnetic latching relay; the second-level identification unit adds a sweep frequency signal generator, an isolation coupling unit, and a synchronous sampling unit; the third-level identification unit adds a wireless mesh communication or power line carrier communication module and supports node encoding and synchronous monitoring. The thermal feature acquisition submodule can use an NTC thermistor, and the acoustic feature acquisition submodule can use a piezoelectric ceramic sheet. All of these devices are low-cost and commonly used, suitable for large-scale deployment on campus.

[0077] The intelligent management and control platform can be deployed on a local server within the school or in the cloud. The platform displays risk status by building, floor, room, and node number. For high-power unauthorized loads, the platform displays the power outage result; for abnormal line topology, the platform displays the abnormal coupling edges and suggested troubleshooting paths; for thermal anomalies and acoustic arcing events, the platform displays the event occurrence time and associated nodes. The platform can also generate risk heat maps for scheduling dormitory inspections and line maintenance.

[0078] Example 8: Complete Identification Process Reference Appendix Figure 8 This embodiment provides a complete identification process.

[0079] Step S0, Initialization. The system assigns a node address and node code to each detection node, completes clock synchronization, baseline spectrum reading, sensor self-test, and safety threshold loading. If a node fails its self-test, it enters degraded mode and reports a maintenance prompt to the intelligent management platform.

[0080] Step S1, Level 1 Power Identification. The Level 1 identification unit collects active power and current in real time. If the active power or current exceeds a preset threshold and continues to exceed a preset time threshold, the system determines it as a high-power non-compliant load, executes power cut-off, and reports an alarm. If the threshold is not exceeded, the system continues to the next level of detection.

[0081] Step S2, Differential Equivalent Impedance Spectrum Scan. The secondary or tertiary identification unit injects a low-amplitude differential frequency sweep signal into the LN loop to acquire the load connection response and node baseline response, calculate the differential equivalent impedance spectrum, and extract impedance spectrum features. If a plugging / unplugging or abnormal grid fluctuation is detected, the frequency sweep is paused.

[0082] Step S3, Coordinated coded frequency sweep detection. For key ports or suspected abnormal line nodes, the three-level identification unit acts as the initiating node, injecting a frequency sweep signal with node coding. Adjacent nodes acquire coupling responses within the synchronous monitoring window and transmit them back. The system confirms the source of the coupling signal through matched filtering.

[0083] Step S4, Coupled Spectrum Topology Determination. The system constructs a coupling spectrum matrix based on the coupling responses between multiple nodes and combines this with the original campus power distribution topology to determine whether there are abnormal coupling edges. If continuous high coupling occurs across circuits or floors, the system outputs a risk of abnormal line topology.

[0084] Step S5: Multi-Physical Domain Feature Acquisition. The system synchronously acquires electrical, thermal, and acoustic features, aligning them with impedance spectral and topological features by timestamp. If a channel experiences a self-test anomaly, the corresponding reliability decrease is recorded.

[0085] Step S6, Reliability Gated Fusion Judgment. The system calculates the reliability of each physical domain, generates fusion weights, inputs the fusion features into the classification model, and combines the physical rule fallback logic to output the load type, violation probability, and line risk level.

[0086] Step S7, Intelligent Management and Control. The intelligent management and control platform performs tiered processing based on the judgment results: power is cut off for high-power resistive loads; warnings are sent and recorded for substandard or abnormal charging loads; troubleshooting work orders are generated for abnormal line topology; on-site prompts and administrator notifications are provided for high temperature rise or arcing acoustic events; and the operating status is only recorded for compliant charging loads that are confirmed to be normal.

Claims

1. A campus lithium battery charging load identification system based on differential equivalent impedance spectroscopy and multi-physics domain intelligent sensing, characterized in that, include: The three-level identification layer includes a first-level identification unit, a second-level identification unit, and a third-level identification unit. The first-level identification unit is deployed at the front end of the socket and is configured with a power threshold detection function. The second-level identification unit is deployed in the centralized charging area and is configured with a differential equivalent impedance spectrum scanning function. The third-level identification unit is deployed at key charging ports and is configured with a coded frequency sweep collaborative detection function. The differential equivalent impedance spectrum scanning module is used to inject low-amplitude differential frequency sweep signals into the live and neutral circuits of the load through isolated coupling, and obtain the differential equivalent impedance spectrum based on the differential result between the load access state response and the node baseline state response. The coded frequency sweeping collaborative detection network consists of three-level identification units deployed at various detection points interconnected through a wireless mesh network or power line carrier communication. It is used to enable adjacent nodes to synchronously collect coupling responses and construct a multi-node coupling spectrum matrix when any three-level identification unit performs frequency sweeping detection as the initiating node. The multi-physical domain intelligent sensing module includes an electrical feature acquisition submodule, a thermal feature acquisition submodule, an acoustic feature acquisition submodule, and a sensor status self-testing submodule; The reliability-gated fusion judgment unit is used to generate fusion weights based on the reliability of electrical characteristics, differential equivalent impedance spectrum characteristics, thermal characteristics, acoustic characteristics and coupling spectrum topology characteristics, and output load type classification results and line topology anomaly judgment results. The intelligent management and control platform is used to receive the load type classification results and line topology anomaly judgment results, and to perform real-time monitoring, hierarchical alarms, remote power outages, baseline spectrum updates, and risk area display.

2. The campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 1, characterized in that, The differential equivalent impedance spectrum scanning module includes a sweep frequency signal generator, an isolation coupling unit, a current limiting protection unit, a dual-channel synchronous sampling analog-to-digital converter, a baseline switching unit, and an impedance spectrum processing unit. The impedance spectrum processing unit calculates the differential equivalent impedance spectrum based on the complex voltage and current responses under the load connection state and the complex voltage and current responses under the node baseline state, so as to reduce the impact of power grid background, line parasitic parameters, and adjacent load coupling on load identification.

3. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 2, characterized in that, The node baseline status is the real-time baseline when the current node has no load access, the online baseline obtained by switching the built-in reference branch, or the equivalent baseline obtained by compensating for cable length and ambient temperature of the calibration baseline of adjacent idle nodes on the same floor and the same circuit.

4. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 1, characterized in that, The frequency range of the low-amplitude differential sweep signal is 100Hz to 100kHz. The sweep frequency points are distributed according to logarithmic coordinates, with no less than 20 frequency points per decade. The peak-to-peak value of the sweep signal does not exceed 5V and the amplitude of the sweep signal does not exceed 2% of the effective value of the current power supply voltage. When abnormal fluctuations in the circuit voltage, sudden high current, or user plugging / unplugging actions are detected, the differential equivalent impedance spectrum scanning module suspends the sweep frequency detection.

5. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 1, characterized in that, The coded frequency sweeping collaborative detection network configures a unique node code for each level-three identification unit; The initiating node modulates its node code onto the sweep frequency signal. Neighboring nodes identify the coupling response within the synchronization monitoring window using a matched filtering algorithm corresponding to the node code, and transmit the amplitude, phase, and arrival time of the coupling response back to the initiating node or the intelligent control platform.

6. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 5, characterized in that, The coded frequency sweeping collaborative detection network constructs a coupling spectrum matrix based on the coupling response between different nodes. When nodes between different power distribution circuits or different floors exhibit a sustained high coupling spectrum value, or when nodes in the original campus power distribution topology that do not have a direct electrical connection exhibit a sustained coupling response, it is determined that there is a shared neutral wire, illegal cross-connection, unauthorized connection, or abnormal insulation coupling.

7. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 1, characterized in that, The thermal feature acquisition submodule uses a negative temperature coefficient thermistor attached near the copper plate or spring of the socket to collect temperature and calculate the temperature rise rate; the acoustic feature acquisition submodule uses a piezoelectric ceramic plate attached near the inner wall of the socket shell or the spring support structure to collect vibration and acoustic signals, and extracts battery management system relay closing events, high-frequency pulse width modulation howling events, and short-time pulse events of arcing or electric arcing.

8. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 1, characterized in that, The reliability-gated fusion determination unit calculates the reliability in the electrical domain, impedance spectrum domain, thermal domain, acoustic domain, and topological domain, respectively, and weights the corresponding physical domain features based on the reliability. When any physical domain sensor self-test is abnormal, the signal-to-noise ratio is lower than the preset benchmark, or the signal is saturated, the fusion weight of the physical domain feature is reduced.

9. A campus lithium battery charging load identification system based on differential equivalent impedance spectrum and multi-physics domain intelligent sensing according to claim 8, characterized in that, The reliability gating fusion judgment unit adopts a lightweight Transformer encoder or equivalent classification model to map electrical features, differential equivalent impedance spectrum features, thermal features, acoustic features, and coupling spectrum topological features into token vectors, and adds the reliability code to the token vectors for cross-domain fusion judgment; when high temperature rise, ignition acoustic events, or abnormal coupling edges across loops are detected, a safety risk warning is output first if the model classification confidence is lower than a preset threshold.

10. A method for identifying campus lithium battery charging loads based on differential equivalent impedance spectroscopy and multi-physics domain intelligent sensing, characterized in that, Includes the following steps: S1, Level 1 Power Identification: Real-time acquisition of active power and current of the charging circuit. When the active power or current exceeds the preset threshold and continues to exceed the preset time threshold, it is determined to be a high-power non-compliant load and a power-off operation is performed. S2. Differential equivalent impedance spectrum scan: Inject low-amplitude differential sweep frequency signals into the live wire and neutral wire loops of the load, obtain the load access state response and node baseline state response respectively, and obtain the differential equivalent impedance spectrum based on the differential results of the two. S3, Coordinated detection by frequency sweeping with encoding: The three-level identification unit acts as the initiating node to inject a frequency sweeping signal with node encoding. Adjacent nodes collect the coupling response within the synchronous monitoring window and transmit the coupling response data back. S4. Topology Anomaly Detection: Construct a coupling spectrum matrix based on the coupling response between multiple nodes, and determine whether there is shared neutral wire, illegal bridging, unauthorized wiring, or insulation coupling anomalies. S5. Multi-physics domain feature acquisition: Simultaneously acquire electrical features, thermal features, and acoustic features, and align the electrical features, thermal features, acoustic features, differential equivalent impedance spectrum features, and coupled spectrum topology features according to timestamps; S6. Reliability Gated Fusion Judgment: Calculate the reliability of each physical domain feature, generate fusion weights based on the reliability, and output the load type classification results and line topology anomaly judgment results; S7. Intelligent Control: Based on the judgment result of step S6, perform hierarchical alarms, remote power outages, risk area display, and baseline spectrum updates.

Images