Photovoltaic prefabricated substation power system operation safety monitoring system

Abstract

The invention relates to the field of electrical monitoring, and discloses a photovoltaic prefabricated substation power system operation safety monitoring system, which comprises a multi-mode sensing network deployed in a prefabricated substation and used for sensing multi-physical field signals during equipment operation; the disturbance injection control interface is used for connecting a photovoltaic inverter controller; the data acquisition and edge computing gateway is connected with the multi-mode sensing network and the disturbance injection control interface; and the data acquisition and edge computing gateway comprises a real-time monitoring and anomaly detection unit and a closed-loop active detection unit. By introducing a closed-loop active detection unit, after the system judges suspected abnormity, a perturbation control signal with frequency spectrum forming is actively injected into inverter PWM logic, and response gain is calculated, so that a closed-loop confirmation effect on fault authenticity is achieved, and the problems that a traditional passive monitoring system only depends on threshold judgment, and the fault detection efficiency is low are solved. And a real fault and transient interference cannot be effectively distinguished.

Description

Technical Field

[0001] This invention relates to the field of electrical monitoring technology, specifically to a power system operation safety monitoring system for photovoltaic prefabricated substations. Background Technology

[0002] As the core hub of a photovoltaic power generation system, a prefabricated photovoltaic substation integrates key high-voltage equipment such as inverters, transformers, and switchgear. To ensure the long-term safe and stable operation of the system, the status of these devices needs to be monitored online in real time to promptly detect potential insulation defects or mechanical failures.

[0003] Existing monitoring methods typically rely on deploying multiple sensors within the substation, such as ultra-high frequency (UHF) sensors for detecting partial discharge, acoustic emission sensors for monitoring equipment vibration, or conventional instrument transformers for monitoring electrical parameters. These systems generally passively acquire physical signals generated during equipment operation, and when certain characteristics of the signals (such as amplitude or energy) exceed pre-set alarm thresholds, the system determines that a fault has occurred.

[0004] However, the electromagnetic environment inside prefabricated photovoltaic substations is extremely complex. When photovoltaic inverters perform high-frequency switching operations, they generate broadband, high-intensity intrinsic switching noise. This strong background noise severely interferes with actual fault signals (such as early, weak partial discharges). This makes it difficult for traditional passive monitoring methods that rely on fixed thresholds to distinguish between real fault characteristics and transient environmental interference, often leading to false alarms or missed alarms and reducing the reliability of the monitoring system.

[0005] When a system detects an abnormal signal, there is often a lack of effective technical means to further verify whether the anomaly is caused by a real physical defect in the equipment itself, or merely by fluctuations in operating conditions or external noise. Furthermore, accurately locating the fault source after confirming a fault is also a technical challenge. Substations have compact internal structures, and the propagation paths of signals within them (including propagation along conductors, through structures, and spatial radiation) are extremely complex. Traditional location algorithms, if employing simplified propagation models, often fail to meet the precise positioning requirements of maintenance personnel. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a power system operation safety monitoring system for prefabricated photovoltaic substations. It solves the problem that traditional passive monitoring methods are unable to effectively distinguish between real equipment faults and transient environmental interference under the background of strong switching noise interference from photovoltaic inverters, resulting in low alarm reliability.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a photovoltaic prefabricated substation power system operation safety monitoring system, comprising: Multimodal sensor networks are deployed in prefabricated substations to sense multi-physics field signals during equipment operation. Disturbance injection control interface, used to connect to the photovoltaic inverter controller; The data acquisition and edge computing gateway is connected to the multimodal sensor network and the disturbance injection control interface. The data acquisition and edge computing gateway includes: The real-time monitoring and anomaly detection unit is used to continuously process the multi-physics field signal, quantitatively compare the multi-physics field signal with the preset health response profile baseline to calculate the system anomaly score, and when the system anomaly score continuously exceeds the anomaly judgment threshold, it is judged as a suspected anomaly, and then the key feature information that causes the anomaly score to rise is identified and output. The closed-loop active detection unit is used to, upon receiving the suspected anomaly and the key feature information, generate a perturbation control signal with specific spectral energy enhancement based on the abnormal frequency band indicated by the key feature information, and inject the perturbation control signal into the pulse width modulation logic unit of the photovoltaic inverter through the perturbation injection control interface, evaluate the response gain of the multiphysics signal to the perturbation control signal, and confirm the suspected anomaly as a confirmed fault and interference based on the response gain.

[0008] Preferably, the real-time monitoring and anomaly detection unit is specifically configured as follows: A rolling time window mechanism is used to generate response profiles that represent the current operating status of the device in real time; Calculate the Mahalanobis distance between the response profile and the preset health response profile baseline, and use the Mahalanobis distance as the system anomaly score; The health response profile baseline includes the mean and covariance matrix of the multidimensional feature vectors extracted under healthy operating conditions.

[0009] Preferably, in the closed-loop active detection unit, the specific method for generating the perturbation control signal with enhanced specific spectral energy includes: If the abnormal frequency band is a continuous frequency range, then a linear frequency modulation signal is generated; If the abnormal frequency band is a set of discrete characteristic frequency points, then a multi-frequency sinusoidal signal is generated; Generate a broadband pseudo-random sequence and pass it through a digital bandpass filter whose passband matches the anomalous frequency band.

[0010] Preferably, in the closed-loop active detection unit, the specific method for evaluating the response gain of the multi-physics signal to the perturbation control signal is as follows: The perturbation control signal is superimposed on the original pulse width modulation control signal of the photovoltaic inverter to form a new pulse width modulation control signal that has been finely tuned. The ratio between the deviation of the abnormal feature detected during active detection and the deviation of the abnormal feature before the detection was triggered is calculated, and the ratio is used as the response gain.

[0011] Preferably, the data acquisition and edge computing gateway is further deployed with: The dual baseline calibration unit is used to collect multi-physics field signals under healthy operating conditions during the system initialization phase, and select one signal as the reference signal and the remaining signals as the response signals. By calculating the normalized cross-correlation function and system transfer function between the reference signal and the response signal, a multi-dimensional feature vector of peak correlation coefficient, peak delay time and transfer function spectral characteristics is extracted to construct the healthy response profile baseline.

[0012] Preferably, the dual baseline calibration unit is also used to calibrate and store the multimodal signal propagation timing topology during the system initialization phase; The specific method for calibrating and storing the multimodal signal propagation timing topology is as follows: Trigger or utilize a transient event from a calibration source at a known location; The signal wavefront of the transient event of the calibration source is recorded with high precision, and the absolute timestamp of its arrival at each sensor is recorded. The difference in timestamps between any two sensors is calculated to construct a time-series topology matrix as the time-series topology diagram for the propagation of the multimodal signal.

[0013] Preferably, the system further includes: The synchronization time unit is used to provide a unified time reference for the data acquisition channels of the multimodal sensor network; The data acquisition and edge computing gateway is also deployed with: The fault source localization unit is used to extract the absolute timestamps of the transient pulse signals of the confirmed fault reaching each sensor in the multimodal sensing network after the closed-loop active detection unit outputs the confirmed fault, and calculate the three-dimensional physical coordinates of the fault source by combining the multimodal signal propagation time sequence topology diagram.

[0014] Preferably, in the fault source localization unit calculation, the specific method for calculating the three-dimensional physical coordinates of the fault source is as follows: A nonlinear optimization objective function is constructed, which is defined as minimizing the sum of squared residuals between the measured arrival time difference of the fault pulse and the propagation time difference predicted based on the propagation time function contained in the multimodal signal propagation time sequence topology. The three-dimensional physical coordinates of the fault source are obtained by solving the nonlinear optimization objective function through an iterative optimization algorithm.

[0015] Preferably, the data acquisition and edge computing gateway is further deployed with: The integrated diagnostic and human-machine interaction module is used to integrate confirmed faults, response gains, and three-dimensional physical coordinates to generate a structured diagnostic report; and is configured to mark the three-dimensional physical coordinates on the three-dimensional digital model or two-dimensional planar layout of the prefabricated substation.

[0016] Preferably, the data acquisition and edge computing gateway is further deployed with: The data preprocessing module is used to align the multi-physics field signal according to a high-precision timestamp, apply digital bandpass filtering, and perform Z-score normalization before the real-time monitoring and anomaly detection unit processes the multi-physics field signal.

[0017] This invention provides a power system operation safety monitoring system for prefabricated photovoltaic substations. It has the following beneficial effects: 1. This invention introduces a closed-loop active detection unit. After the system determines a suspected abnormality, it actively injects a spectrum-shaping perturbation control signal into the inverter PWM logic and calculates the response gain. This achieves a closed-loop confirmation effect of the authenticity of the fault, solving the technical problem that traditional passive monitoring systems rely solely on threshold discrimination and cannot effectively distinguish between real faults and transient interference. This results in a significant improvement in alarm reliability and a significant reduction in false alarm rate.

[0018] 2. This invention constructs a health response profile baseline based on endogenous switching noise and combines it with the Mahalanobis distance algorithm to calculate the comprehensive anomaly score of multidimensional features, thereby achieving high sensitivity detection of early weak state deviations of the system. This solves the technical problem of difficulty in identifying early fault budding features under strong switching noise background and significantly improves the ability to detect faults early.

[0019] 3. This invention constructs a multimodal signal propagation time-series topology map using known calibration sources during the calibration phase, and integrates it as prior knowledge into the nonlinear optimization model for fault source localization. This achieves high-precision calculation of the physical coordinates of the fault source, solves the technical problem of large positioning errors caused by traditional positioning algorithms ignoring complex propagation paths within substations, and provides maintenance personnel with more accurate fault location directions. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is an architecture diagram of the data acquisition and edge computing gateway module of the present invention. Detailed Implementation

[0021] See attached document Figure 1 , Figure 1This is a schematic diagram of a safety monitoring system architecture for a pre-installed photovoltaic substation according to an embodiment of the present invention. The present invention provides a safety monitoring system, which may include: a multimodal sensor network 110, a synchronization and timing unit 120, a data acquisition and edge computing gateway 130, and a disturbance injection control interface 140.

[0022] A multimodal sensor network 110, deployed at monitoring points within a prefabricated substation, is used to sense multi-physics field signals during equipment operation. This network includes: High-frequency electrical sensors used for acquiring high-frequency electrical signals; Acoustic emission sensor used to collect ultrasonic signals from equipment structures; Ultra-high frequency sensors used to collect electromagnetic signals of insulation discharge.

[0023] The synchronization and timing unit 120 provides a unified time reference for all data acquisition channels within the system. It ensures sub-microsecond synchronization of the signal data acquired by the multimodal sensor network 110 through a precise time protocol or satellite time synchronization. The data acquisition and edge computing gateway 130, connected to the multimodal sensor network 110 and the synchronization and timing unit 120, is used to collect and process data and integrates a high-speed data acquisition channel and data processor execution logic.

[0024] The disturbance injection control interface 140 connects the data acquisition and edge computing gateway 130 and the photovoltaic inverter controller in the substation. This interface is used to transmit the digital control commands generated by the data acquisition and edge computing gateway 130 to the pulse width modulation (PWM) logic unit of the inverter.

[0025] See attached document Figure 2 , Figure 2 This is a system functional module block diagram according to an embodiment of the present invention. The functional modules deployed on the data acquisition and edge computing gateway 130 may include: a data preprocessing module 210, a dual baseline calibration unit 220, a real-time monitoring and anomaly detection unit 230, a closed-loop active detection unit 240, a fault source localization unit 250, and a comprehensive diagnosis and human-machine interaction module 260.

[0026] The overall workflow of the monitoring method disclosed in this invention is as follows: During the system initialization phase, the operator initiates the calibration procedure. The dual-baseline calibration unit 220 first acquires the intrinsic switching noise signal generated by the inverter and its response signal on the multimodal sensor network 110 under healthy operating conditions, and then calculates and stores a health response profile baseline. Subsequently, the dual-baseline calibration unit 220 establishes and stores a multimodal signal propagation time-series topology by recording the propagation time difference of a preset transient event with a known location on each sensor.

[0027] During the normal operation phase of the system, the real-time monitoring and anomaly detection unit 230 continuously collects and processes data from the multimodal sensor network 110. This unit calculates a response profile representing the current device status in real time and compares it quantitatively with the stored health response profile baseline to obtain a system anomaly score.

[0028] When the system anomaly score calculated by the real-time monitoring and anomaly detection unit 230 continuously exceeds a preset threshold, it determines that a suspected anomaly has occurred in the system. At this time, the unit identifies the key feature information that caused the anomaly score to rise and transmits this information, along with a trigger command, to the closed-loop active detection unit 240.

[0029] Upon receiving a trigger command, the closed-loop active detection unit 240 generates a perturbation control signal with enhanced specific spectral energy based on the input key feature information. This signal is then used to fine-tune the inverter's PWM control logic via the perturbation injection control interface 140, thereby actively shaping the spectrum of perturbation sources within the system.

[0030] While actively applying a disturbance, the closed-loop active detection unit 240 analyzes the system's response to this enhanced disturbance. The response gain for specific characteristics is evaluated. If the gain is significant, the suspected anomaly is confirmed as a real fault; If the gain is not significant, it is determined to be external interference.

[0031] For a genuine fault identified as having transient pulse characteristics, the system triggers the fault source localization unit 250. This unit extracts the precise timestamps of the fault pulse signals arriving at each sensor and, in conjunction with the calibrated multimodal signal propagation timing topology, calculates the three-dimensional physical coordinates of the fault source using a nonlinear optimization algorithm.

[0032] Finally, the comprehensive diagnosis and human-computer interaction module 260 integrates the analysis results such as the nature of the fault, the confirmation status, and the physical location to generate a structured diagnostic report for display to maintenance personnel.

[0033] See attached document Figure 1 With appendix Figure 2 The data acquisition and edge computing gateway 130 is connected to the multimodal sensor network 110 via a high-speed data bus and receives time reference signals from the synchronization and timing unit 120. The data preprocessing module 210 deployed within this gateway is implemented as follows: S4211: The system performs synchronous acquisition and transmission of multi-channel data. Each sensor in the multimodal sensor network 110, while acquiring physical signals and converting them into digital signals, is given a high-precision timestamp by the synchronization timing unit 120. This timestamped data is transmitted in the form of data packets to the data acquisition and edge computing gateway 130 via communication methods such as industrial Ethernet.

[0034] S4212: After receiving data packets from different sensor channels, the data preprocessing module 210 first aligns the data in its internal data buffer. Based on the timestamp information within each data packet, this module sorts and interpolates the data from different channels to reconstruct a fully aligned multidimensional data matrix along the time axis. Each row of this matrix corresponds to a sampling time, and each column corresponds to a sensor channel, thus ensuring data consistency for subsequent cross-modal analysis.

[0035] S4213: The data preprocessing module 210 applies a digital bandpass filter to each channel of data in the time-aligned data matrix. The purpose of this filtering is to suppress frequency band interference unrelated to the features being analyzed. Specifically, the passband range of the digital bandpass filter is set to cover the main frequency band of the photovoltaic inverter's switching noise and the characteristic frequency band of typical equipment fault signals (such as partial discharge), for example, in the kHz to MHz range. It can filter out interference from the 50Hz power frequency and its lower harmonics, and also filter out high-frequency white noise that exceeds the effective response range of the sensor. For the specific implementation of the digital filter, those skilled in the art can design it using IIR or FIR filters; the design methods are well-known in the field and will not be elaborated upon here.

[0036] S4214: To eliminate signal amplitude differences between different sensor channels due to variations in their gain, sensitivity, and physical installation location, the data preprocessing module 210 normalizes the filtered data. One specific implementation method is to use Z-score normalization. For the signal time series of any channel... Its normalized signal The calculation method is as follows: ; In the formula, Index for discrete time points; This represents the average value of the channel signal within the rolling time window. This represents the standard deviation of the channel signal within the same rolling time window. The signal is normalized; For any channel, the signal time series; This is the average value of the channel signal over a rolling time window.

[0037] This process converts the signals from all channels to a uniform dimensionless scale, facilitating fair feature extraction and model comparison in the future.

[0038] After the above steps, the data preprocessing module 210 outputs a multidimensional data stream with a clean time-synchronized frequency band and normalized amplitude, which is then supplied to the dual baseline calibration unit 220 or the real-time monitoring and anomaly detection unit 230 for subsequent in-depth analysis.

[0039] See attached document Figure 2 In system initialization calibration mode, the dual baseline calibration unit 220 performs the operation of constructing a health response profile baseline. This unit receives a data stream characterizing the device's health status from the data preprocessing module 210, and its specific implementation is as follows: S4221: The dual-baseline calibration unit 220 selects one signal from the preprocessed data stream as the reference signal for system excitation, denoted as the discrete-time sequence. This reference signal is typically selected from monitoring points that can stably characterize the switching behavior of the photovoltaic inverter, such as a high-frequency current signal from the AC bus. The rest... Signals from different monitoring points are used as the system's response signals, denoted as... ,in 。; S4222: To quantize the transfer characteristics from the excitation source to each response point in the time domain, this unit calculates the reference signal. With each response signal Normalized cross-correlation function between This function characterizes the similarity between two signals at different time delays, and its calculation formula is as follows: In the formula, Index for discrete time points; This is the time delay step size; For reference signal and the first The response signal has a time delay Normalized cross-correlation function at the location; This represents the mean of the corresponding signal sequence within the calculation window; For reference signal; In response to the signal; The mean of the reference signal within the calculation window; For the first The response signal at discrete time points The value of . S4223: To quantize the system transfer characteristics in the frequency domain, this element estimates the value from the reference point to the . System transfer function at each response point One specific implementation method is to calculate the power spectral density of the signal.

[0040] First, calculate the self-power spectral density of the reference signal. Cross-power spectral density of the reference signal and the response signal Then the transfer function can be estimated as: ; In the formula, For frequency; The system transfer function; The auto-power spectral density of the reference signal; The cross-power spectral density is the cross-power spectral density of the reference signal and the response signal. For calculating the power spectral density, those skilled in the art can use methods such as the Welch method based on the Fast Fourier Transform (FFT). S4224: The dual-baseline calibration unit 220 extracts a set of quantitative features that can comprehensively characterize the health status of the equipment from the time-domain and frequency-domain characteristics obtained above, as well as from the original signal. These features are combined into a multi-dimensional feature vector. As the first The digital fingerprint of each monitoring point in its healthy state. This feature vector It may include the following components: Peak correlation coefficient Take the cross-correlation function The maximum value of represents the fidelity of signal transmission.

[0041] Peak latency Time delay when the cross-correlation function reaches its peak ( The sampling time interval; The group delay (the specific number of sampling points that delays the signal propagation to maximize the correlation) characterizes the group delay of the signal propagation.

[0042] Response energy ratio : Calculate the ratio of the response signal energy to the reference signal energy to characterize the energy attenuation during signal transmission.

[0043] Transfer function spectral features: Extracting transfer function At one or more key frequencies amplitude at With phase Alternatively, parameters such as the center frequency and quality factor of its resonance peak can be extracted.

[0044] S4225: Dual baseline calibration unit 220 will... Feature vector of each monitoring point The data is aggregated to form a system-level health response profile baseline. Simultaneously, this unit also calculates the value of each feature vector under healthy operating conditions. The statistical distribution characteristics of each internal component, especially its covariance matrix Ultimately, this includes the mean of the feature vectors (i.e. and covariance matrix Complete health response profile baseline It is stored in the non-volatile memory of the data acquisition and edge computing gateway 130 as a reference for subsequent state comparison.

[0045] See attached document Figure 2 The dual-baseline calibration unit 220, while constructing the health response profile baseline, also performs the operation of constructing a multimodal signal propagation time-series topology map. The purpose of this operation is to calibrate a benchmark model characterizing the signal propagation characteristics in the substation's physical structure for subsequent fault source localization. Its specific implementation is as follows: S4226: In system initialization calibration mode, the dual-baseline calibration unit 220 triggers or utilizes a system-level broadband transient event at a known location as a calibration source. One specific implementation involves controlling a circuit breaker (e.g., the main incoming line circuit breaker) within the prefabricated substation to be at a determined location. Perform a tripping or closing operation. This operation can simultaneously generate signals such as electromagnetic transient pulses and mechanical vibration sound waves, which can be clearly captured by various types of sensors in the multimodal sensor network 110.

[0046] S4227: The data acquisition and edge computing gateway 130 uses the time reference provided by the synchronization timing unit 120 to record the transient signal wavefront generated by the calibration source event with high precision, and delivers it to each sensor in the multimodal sensor network 110. The absolute timestamp, denoted as For accurate extraction of wavefront arrival time, a threshold method based on instantaneous changes in signal energy can be used, or an algorithm such as the Akaike Information Criterion can be used for automatic calibration to identify the precise moment of signal abrupt change.

[0047] S4228: Dual Baseline Calibration Unit 220 (All) A set of arrival timestamps recorded by each sensor Based on this, a multimodal signal propagation time-series topology diagram is constructed. One specific implementation involves constructing this topology diagram as a time-series topology matrix. The elements of the matrix Defined as the calibration signal arriving at the sensor and sensors Time difference between them: In the formula, and This is the sensor index, with values ​​ranging from 1 to... ; and These are the absolute timestamps of the calibration event signals arriving at the i-th and j-th sensors, respectively.

[0048] S4229: This time-series topology matrix It is stored in the non-volatile memory of the data acquisition and edge computing gateway 130. This matrix... Quantification from the same source point (i.e. The signal emitted by the device travels through the complex physical propagation paths within the substation (e.g., electrical signals propagating along bus conductors, acoustic signals propagating through the air, and electromagnetic signals passing through insulating media) to reach any two sensors. and The actual propagation time difference between them. This matrix This will serve as prior knowledge characterizing the system's physical topology and signal propagation characteristics, which will be used by the subsequent fault source location unit 250 when performing location calculations.

[0049] See attached document Figure 2 The real-time monitoring and anomaly detection unit 230 is activated during normal system operation and continuously performs monitoring tasks. This unit receives real-time data streams from the data preprocessing module 210 and invokes the health response profile baseline established by the dual baseline calibration unit 220. The specific implementation method is as follows: S4231: The real-time monitoring and anomaly detection unit 230 employs a rolling time window mechanism to segment the continuous multi-channel data stream from the data preprocessing module 210. For the data within each time window, the real-time monitoring and anomaly detection unit 230 repeatedly executes all calculation steps in steps S4221 to S4224, including calculating the cross-correlation function, estimating the transfer function, and extracting multi-dimensional features. Thus, the system generates a response profile characterizing the current operating status of the equipment in real time. The image is composed of the feature vector set at the current moment. constitute.

[0050] S4232: The real-time monitoring and anomaly detection unit 230 will generate a current response profile in real time. Baseline of storage health response profile Perform quantitative comparisons to calculate the current comprehensive anomaly score of the system. To improve the sensitivity of detection to early, minor deviations while suppressing normal data drift caused by fluctuations in operating conditions during equipment operation in a healthy state, the calculation of the anomaly score preferably uses Mahalanobis distance. This distance utilizes the health state covariance matrix stored in step S4225. This takes into account the statistical correlation between features. Among them, the system's comprehensive anomaly analysis... It can be given by the following formula: ;

[0051] In the formula, Indicates the current time; The index for the monitoring points, from 1 to ; For the preset first The weight coefficient of each monitoring point can be set according to its importance, or uniformly set to 1; For the first Each monitoring point at the current time eigenvectors; For storage of the first Health baseline feature vectors of each monitoring point; For the first Health status covariance matrix of monitoring points The inverse matrix.

[0052] S4233: Real-time monitoring and anomaly detection unit 230 calculates the comprehensive anomaly distribution in real time. Compared with the preset anomaly detection threshold Compare. The threshold. During the calibration phase, health data can be used as a basis (e.g., taking data from a healthy state). Distribution The settings can be determined by quantiles or dynamically adjusted using an adaptive algorithm. To prevent false alarms caused by transient interference, the system can be configured with a duration judgment logic, i.e., when... continuous Number of sampling points (or duration) All are greater than the threshold. Only then does the system officially determine it as a suspected abnormal state.

[0053] S4234: After determining a suspected anomaly, the real-time monitoring and anomaly detection unit 230 does not simply issue an alarm, but further performs anomaly feature identification. This unit analyzes... The calculation process traces back to the component that contributed the most to the anomaly. Specifically, it can identify which one or more monitoring points contributed to the anomaly. Mahalanobis distance The largest increase occurred, and the feature vector at that monitoring point was [data missing]. In the middle, which one or more specific feature components (e.g., a specific frequency band) are being referred to? The magnitude of the transfer function deviated from its reference value. Most of the information identified (e.g., specific sensor location i and anomalous frequency bands) The key feature information is combined and sent to the closed-loop active detection unit 240 along with the trigger command as the basis for the next active detection step.

[0054] See attached document Figure 2 The closed-loop active detection unit 240, deployed within the data acquisition and edge computing gateway 130, is activated upon receiving a suspected anomaly trigger command from the real-time monitoring and anomaly detection unit 230. The specific implementation of this unit is as follows: S4241: The closed-loop active detection unit 240 first parses the trigger command received from the real-time monitoring and anomaly detection unit 230. This command contains key feature information, which identifies one or more abnormal frequency bands (causing an increase in the system's abnormal score) identified in step S4234, denoted as... .

[0055] S4242: One of the core functions of the closed-loop active detection unit 240 is to detect abnormal frequency bands. Dynamically design and generate digital perturbation control signals. This process is called perturbation spectrum shaping. The signal... The spectral energy is intentionally concentrated in the target anomalous frequency band. Internally, energy in other frequency bands (especially power frequency and its lower harmonic bands) is strictly suppressed to ensure that this active detection behavior does not cause observable disturbances to the normal power transmission of the substation.

[0056] S4243: To achieve the above-mentioned spectrum shaping, this invention provides several specific implementation methods. One implementation method is to generate a linear frequency modulated signal. If an abnormal frequency band is identified... It is a continuous frequency range Then, within the preset detection period Inside, the perturbation control signal It can be given by the following formula: ; In the formula, This serves as a reference amplitude for controlling the injected energy; Index for discrete time points; The digital sampling time interval of the system; and These are the start and end frequencies of the frequency sweep, respectively. This signal can excite a complete sweep of the target frequency band. For the detection period; For discrete time points The perturbation control signal value.

[0057] Another way to implement it is, if It manifests as discrete characteristic frequency points Then the perturbation control signal These can be synthesized as a superposition of sinusoidal signals of these frequencies, i.e., a multi-frequency sinusoidal signal: ; In the formula, and They are the first The amplitude and initial phase of each frequency component. This is achieved by appropriately setting the phase of each component. (For example, using Schroder phase) can reduce the peak-to-average power ratio of the signal; For discrete time points The value of the perturbation control signal; The digital sampling time interval of the system; The index of the currently superimposed sinusoidal signal component, from 1 to... m represents the total number of sinusoidal signal components that need to be superimposed and synthesized, corresponding to the number of discrete characteristic frequency points.

[0058] Another implementation involves generating a wideband pseudo-random sequence (PRBS) and passing it through a digital bandpass filter. The passband of this digital bandpass filter is precisely designed to correspond to the anomalous frequency band. The filters are matched. After this filtering process, the energy of the pseudo-random sequence is shaped and concentrated within the target frequency band. For the design of digital filters, those skilled in the art can use FIR or IIR filters, which will not be elaborated here.

[0059] S4244: The closed-loop active detection unit 240 generates the perturbation control signal with completed spectrum shaping in the above steps. The output is used for subsequent perturbation injection steps.

[0060] See attached document Figure 1 With appendix Figure 2 In step S4244, the closed-loop active detection unit 240 generates a perturbation control signal. Next, the disturbance injection control and response gain evaluation steps are performed. The specific implementation is as follows: S4245: The closed-loop active detection unit 240 will generate the digital perturbation control signal in step S4244. The disturbance signal is sent to the PWM controller of the photovoltaic inverter via the disturbance injection control interface 140. Inside this controller, the disturbance signal interacts with the inverter's existing PWM control signal. The signals are superimposed to form a new PWM control signal that is finely tuned. : In the formula, The original PWM control signal sequence of the inverter under the current operating conditions; at discrete time points The value of the perturbation control signal; It is a tiny gain coefficient; This is the new PWM control signal to be fine-tuned. This coefficient... The value of is strictly limited to ensure that the actively injected disturbance energy is controlled at an extremely low level, thereby achieving excitation enhancement of a specific frequency band without affecting the normal power conversion function and grid friendliness of the inverter.

[0061] S4246: Inverter with new PWM control signal During operation, the spectral characteristics of the excitation source within the system were in abnormal frequency bands. Active enhancement is obtained. At this time, the closed-loop active detection unit 240 synchronously controls the multimodal sensor network 110 and the data preprocessing module 210 to collect and process the system response signal of the active detection stage in the same way as in step S4231, and recalculate the feature components corresponding to the key abnormal features identified in step S4234.

[0062] S4247: The closed-loop active detection unit 240 quantifies and evaluates the feature changes before and after active detection, i.e., calculates the response gain. The gain Defined as during active detection Deviation of detected abnormal features Before the detection was triggered Deviation of this feature The ratio between them: ; In the formula, This is an index of the key abnormal features identified in step S4234; This feature is used to classify anomalies before triggering active detection. The contribution value (or its physical deviation); The contribution value (or physical deviation) of this feature under the same calculation method after active stimulation is applied.

[0063] S4248: The closed-loop active detection unit 240 uses the calculated response gain Make a final fault confirmation decision. One specific implementation is to adjust the response gain. With a preset gain threshold ( Typically, a constant significantly greater than 1 (e.g., 5 or 10) is used for comparison. This indicates that after applying a weak excitation to the suspected abnormal frequency band, the system's response in that band showed a much larger amplification than proportional. This confirms that the system characteristics in that frequency band have undergone a real physical change (e.g., the generation of a new mechanical or electrical resonance point). Therefore, the system confirms the suspected abnormality identified in step S4233 as a confirmed fault. If the system response does not increase proportionally with the stimulus, the determination in step S4233 is likely caused by transient environmental interference or random noise. The system classifies this event as interference and cancels the warning status. This confirmation result (fault confirmed or interference cleared) is sent to the integrated diagnostic and human-machine interaction module 260. If the result is a confirmed fault, the instruction will also be sent synchronously to the fault source location unit 250 to initiate the subsequent location process.

[0064] See attached document Figure 2 The fault source location unit 250 is activated after receiving a confirmed fault trigger command from the closed-loop active detection unit 240, targeting a fault with transient pulse characteristics (such as partial discharge). Its specific implementation is as follows: S4251: The fault source localization unit 250 retrieves the high-sampling-rate raw multi-channel data stream at the moment the confirmed fault event occurred from the data buffer of the data acquisition and edge computing gateway 130. This unit applies a wavefront arrival time extraction algorithm to each of the N sensor channels in the multimodal sensor network 110 to accurately detect the absolute arrival timestamp of the fault transient pulse's wavefront, denoted as... ,in For accurate extraction of wavefront arrival time, algorithms based on signal energy, wavelet transform, or Akaike Information Criterion (AIC) can be used to identify the precise moment of signal abrupt change.

[0065] S4252: The core task of this unit is to construct a nonlinear optimization objective function and determine the unknown physical coordinates of the fault source by solving this function. The objective function It is constructed as the arrival time difference of each sensor relative to a reference sensor (e.g., sensor 1), as measured in step S4251. The baseline model, which characterizes the signal propagation characteristics in the substation physical structure and is labeled by the multimodal signal propagation time-series topology diagram constructed by the dual baseline calibration unit 220, is used to predict the propagation characteristics of the signal from the assumed fault point. Time difference of propagation to each sensor The sum of squared residuals between them is minimized. Its mathematical expression is: ; In the formula, Let be the value of the objective function to be minimized; For the sensor index, traverse from 2 to ; and The fault pulses measured in step S4251 reach the first... The absolute timestamps of the first sensor and the first (reference) sensor; The three-dimensional coordinates of the fault source to be solved ; and The first The known three-dimensional coordinates of the first sensor and the first sensor; To characterize the signal from the source point Spread to monitoring points The propagation time function of the required time.

[0066] S4253: In the formula of step S4252, the propagation time function The construction incorporates the multimodal signal propagation timing topology matrix established in step S4229. The prior knowledge included. The propagation time function of this function. Instead of employing a simple linear model based on a single and constant propagation speed (i.e., distance / velocity), a nonlinear propagation model was constructed. This model utilizes the temporal topology matrix MT, provided by the matrix for known points, during the calibration phase. The actual propagation time difference data is calibrated to establish an accurate mapping relationship that reflects the complex physical structure, electromagnetic environment, and mixed propagation path characteristics of multimodal signals (electric, acoustic, and electromagnetic) within the substation.

[0067] S4254: The fault source location unit 250 finds the fault source by solving the nonlinear least squares optimization problem in S4252. Optimal coordinates to reach the minimum value One specific implementation method is to use iterative optimization algorithms, such as the Gauss-Newton method or the Levenberg-Marquardt method, to optimize the objective function. The solution is obtained through iterative steps. The implementation of such nonlinear optimization algorithms is well-known in this field and will not be elaborated upon here.

[0068] S4255: The optimal solution obtained after the iterative solution process finally converges. This refers to the three-dimensional physical coordinates of the fault source output by this invention. The fault source localization unit 250 sends this coordinate result to the integrated diagnosis and human-computer interaction module 260 for final report generation.

[0069] See attached document Figure 2 The integrated diagnostic and human-computer interaction module 260 is deployed on the data acquisition and edge computing gateway 130, serving as the system's final information output and interaction unit. Its specific implementation is as follows: S4261: The integrated diagnostic and human-machine interaction module 260 is configured to receive processing results from other functional units. These results include: system anomaly analysis data transmitted by the real-time monitoring and anomaly detection unit 230. and key anomaly features The fault confirmation status (i.e., fault or interference has been confirmed) and response gain transmitted by the closed-loop active detection unit 240 ; and the physical coordinates of the fault transmitted by the fault source location unit 250. .

[0070] S4262: This unit integrates and formats the received information to generate a structured diagnostic report. This report contains a multi-field data structure to comprehensively describe the confirmed fault events.

[0071] The structured diagnostic report includes at least the following fields: a unique event identifier; a timestamp of the event occurrence; the nature of the fault identified by the real-time monitoring and anomaly detection unit 230 (e.g., transient pulse characteristics or specific frequency band resonance); the confirmation status given by the closed-loop active detection unit 240; and the response gain used to support this confirmation. Quantized values; physical coordinates of the fault calculated by the fault source location unit 250. In addition, the report may include an index pointing to the raw waveform data related to the event stored in the data acquisition and edge computing gateway 130 for operation and maintenance personnel to access and review.

[0072] S4263: The integrated diagnostics and human-computer interaction module 260 is also responsible for providing a human-computer interaction interface to maintenance personnel. This interface includes an event log list, which displays the structured diagnostic report generated in step S4262 in chronological order.

[0073] The human-machine interface also includes a graphical user interface (GUI). This GUI is pre-loaded with a two-dimensional plan layout or a three-dimensional digital model of the prefabricated substation. When maintenance personnel select a diagnostic report containing valid coordinates from the event log list, the GUI will automatically retrieve the physical coordinates of the fault from the report. The precise physical location of the fault source is then highlighted as an icon on the 2D layout diagram or 3D model. Maintenance personnel can then select this highlighted icon to access a detailed, structured diagnostic report corresponding to that location.

[0074] S4264: In addition, the integrated diagnostic and human-machine interaction module 260 will also send the generated structured diagnostic report to the upper-level monitoring system (such as SCADA) or the local historical database for long-term archiving, statistical analysis and trend prediction of fault events.

Claims

1. A photovoltaic prefabricated substation power system operation safety monitoring system, characterized in that, include: Multimodal sensor networks are deployed in prefabricated substations to sense multi-physics field signals during equipment operation. Disturbance injection control interface, used to connect to the photovoltaic inverter controller; The data acquisition and edge computing gateway is connected to the multimodal sensor network and the disturbance injection control interface. The data acquisition and edge computing gateway includes: The real-time monitoring and anomaly detection unit is used to continuously process the multi-physics field signal, quantitatively compare the multi-physics field signal with the preset health response profile baseline to calculate the system anomaly score, and when the system anomaly score continuously exceeds the anomaly judgment threshold, it is judged as a suspected anomaly, and then the key feature information that causes the anomaly score to rise is identified and output. The closed-loop active detection unit is used to, upon receiving the suspected anomaly and the key feature information, generate a perturbation control signal with specific spectral energy enhancement based on the abnormal frequency band indicated by the key feature information, and inject the perturbation control signal into the pulse width modulation logic unit of the photovoltaic inverter through the perturbation injection control interface, evaluate the response gain of the multiphysics signal to the perturbation control signal, and confirm the suspected anomaly as a confirmed fault and interference based on the response gain.

2. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 1, characterized in that, The real-time monitoring and anomaly detection unit is specifically configured as follows: A rolling time window mechanism is used to generate response profiles that represent the current operating status of the device in real time; Calculate the Mahalanobis distance between the response profile and the preset health response profile baseline, and use the Mahalanobis distance as the system anomaly score; The health response profile baseline includes the mean and covariance matrix of the multidimensional feature vectors extracted under healthy operating conditions.

3. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 1, characterized in that, The specific methods for generating the perturbation control signal with enhanced specific spectral energy in the closed-loop active detection unit include: If the abnormal frequency band is a continuous frequency range, then a linear frequency modulation signal is generated; If the abnormal frequency band is a set of discrete characteristic frequency points, then a multi-frequency sinusoidal signal is generated; Generate a broadband pseudo-random sequence and pass it through a digital bandpass filter whose passband matches the anomalous frequency band.

4. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 1, characterized in that, In the closed-loop active detection unit, the specific method for evaluating the response gain of the multi-physics field signal to the perturbation control signal is as follows: The perturbation control signal is superimposed on the original pulse width modulation control signal of the photovoltaic inverter to form a new pulse width modulation control signal that has been finely tuned. The ratio between the deviation of the abnormal feature detected during active detection and the deviation of the abnormal feature before the detection was triggered is calculated, and the ratio is used as the response gain.

5. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 1, characterized in that, The data acquisition and edge computing gateway is also deployed with: The dual baseline calibration unit is used to collect multi-physics field signals under healthy operating conditions during the system initialization phase, and select one signal as the reference signal and the remaining signals as the response signals. By calculating the normalized cross-correlation function and system transfer function between the reference signal and the response signal, a multi-dimensional feature vector of peak correlation coefficient, peak delay time and transfer function spectral characteristics is extracted to construct the healthy response profile baseline.

6. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 5, characterized in that, The dual baseline calibration unit is also used to calibrate and store the multimodal signal propagation timing topology during the system initialization phase. The specific method for calibrating and storing the multimodal signal propagation timing topology is as follows: Trigger or utilize a transient event from a calibration source at a known location; The signal wavefront of the transient event of the calibration source is recorded with high precision, and the absolute timestamp of its arrival at each sensor is recorded. The difference in timestamps between any two sensors is calculated to construct a time-series topology matrix as the time-series topology diagram for the propagation of the multimodal signal.

7. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 1, characterized in that, The system also includes: The synchronization time unit is used to provide a unified time reference for the data acquisition channels of the multimodal sensor network; The data acquisition and edge computing gateway is also deployed with: The fault source localization unit is used to extract the absolute timestamps of the transient pulse signals of the confirmed fault reaching each sensor in the multimodal sensing network after the closed-loop active detection unit outputs the confirmed fault, and calculate the three-dimensional physical coordinates of the fault source by combining the multimodal signal propagation time sequence topology diagram.

8. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 7, characterized in that, In the fault source localization unit calculation, the specific method for calculating the three-dimensional physical coordinates of the fault source is as follows: A nonlinear optimization objective function is constructed, which is defined as minimizing the sum of squared residuals between the measured arrival time difference of the fault pulse and the propagation time difference predicted based on the propagation time function contained in the multimodal signal propagation time sequence topology. The three-dimensional physical coordinates of the fault source are obtained by solving the nonlinear optimization objective function through an iterative optimization algorithm.

9. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 8, characterized in that, The data acquisition and edge computing gateway is also deployed with: The integrated diagnostic and human-machine interaction module is used to integrate confirmed faults, response gains, and three-dimensional physical coordinates to generate a structured diagnostic report; and is configured to mark the three-dimensional physical coordinates on the three-dimensional digital model or two-dimensional planar layout of the prefabricated substation.

10. The photovoltaic prefabricated substation power system operation safety monitoring system according to claim 1, characterized in that, The data acquisition and edge computing gateway is also deployed with: The data preprocessing module is used to align the multi-physics field signal according to a high-precision timestamp, apply digital bandpass filtering, and perform Z-score normalization before the real-time monitoring and anomaly detection unit processes the multi-physics field signal.

Images