Subway flexible direct power supply equipment state monitoring method and device of energy router

Abstract

This application provides a method and apparatus for status monitoring of subway flexible DC power supply equipment using an energy router, relating to the field of data processing. Vibration pickup units are arranged on key components of the energy router and time-synchronized. The pickup sequence is aligned with the operating state sequence. Operating segments are divided based on fault ride-through, current limiting, and recovery flags, constructing a structural acoustic spectrum sequence. A steady-state spectrum reference is formed during the steady-state phase. During the recovery phase, attractor regression is used to identify whether the internal state has returned to normal. When deviation persists, implicit recovery is marked as incomplete. Secondary anomalies caused by subsequent load mutations are linked to previous implicit states through spectrum inheritance relationships. Implementing this technical solution facilitates the identification and attribution of implicit recovery failures caused by the incomplete release of slow variables in the control loop, thereby improving the accuracy of status monitoring of subway flexible DC power supply equipment using an energy router.

Description

Technical Field

[0001] This application relates to the technical field of data processing, specifically to a method and device for monitoring the status of a subway flexible DC power supply device using an energy router. Background Technology

[0002] With the continuous expansion of urban rail transit, subway power supply systems are gradually evolving from traditional AC traction power supply systems to flexible DC power supply systems. This aims to improve energy utilization efficiency, enhance regenerative braking energy recovery capabilities, and increase the flexibility and reliability of system operation. Under the flexible DC power supply architecture, the energy router, as a crucial core device connecting traction loads, energy storage units, and the power grid, undertakes key functions such as multi-port energy allocation, voltage stability control, and adaptive adjustment of operating conditions. Its operational status directly affects the safety, stability, and continuity of the subway power supply system.

[0003] In real-world operating environments, metro flexible DC power supply systems frequently need to handle complex conditions such as train starting, braking, section switching, and sudden faults, making voltage drops, current limiting constraints, and short-term disturbances unavoidable. To ensure the safe operation of the system under these conditions, energy routers generally employ multi-layered control structures, including integral regulation, amplitude limiting control, and anti-saturation mechanisms, enabling them to maintain controlled output even under extreme conditions. However, existing metro power supply equipment condition monitoring technologies primarily focus on externally measurable electrical parameters such as voltage, current, and temperature. Their monitoring logic often uses threshold exceedances, steady-state deviations, or short-term anomalies as criteria, lacking effective perception capabilities for latent state changes that evolve slowly within the control system and have little impact on external output. Given the increasingly tight operating rhythm and frequent load fluctuations of flexible DC power supply systems, this monitoring method struggles to promptly reveal the potential risks accumulated by energy routers during fault ride-through and recovery phases, resulting in low accuracy in condition monitoring.

[0004] Therefore, there is an urgent need for a method and device for monitoring the status of flexible DC power supply equipment in subways using an energy router. Summary of the Invention

[0005] This application provides a method and device for monitoring the status of subway flexible DC power supply equipment using an energy router, which facilitates the identification and attribution of implicit recovery failures caused by the incomplete release of slow variables in the control loop, thereby improving the accuracy of status monitoring of subway flexible DC power supply equipment using an energy router.

[0006] The first aspect of this application provides a method for monitoring the status of a subway flexible DC power supply device for an energy router. The energy router's DC-side filter inductor, AC-side parallel reactor, bus thin-film capacitor, and power module heat dissipation base are pre-configured with vibration pickup units, and each vibration pickup unit is bound to a unique installation location identifier and a time synchronization identifier. The method includes: during the operation of the energy router, acquiring a vibration pickup sequence sent by the vibration pickup unit and an operating status sequence bound to the energy router's operation process; and aligning the vibration pickup sequence and the operating status sequence based on the time synchronization identifier to form an aligned sequence. The operating status sequence includes a fault ride-through flag, a current limiting flag, and a recovery flag. Based on the fault ride-through flag, the current limiting flag, and the recovery flag, the aligned sequence is segmented to obtain target time segments. The target time segments include a drop-in segment, a current limiting maintenance segment, a recovery release segment, and a steady-state re-entry segment. Structural acoustic pattern generation processing is performed on the target time segments to obtain pattern segments, and a pattern map is constructed based on the pattern segments to form a pattern map sequence. The sections of the pattern map... Points are represented by peak clusters in the spectral peak set, and the edges of the spectral pattern are represented by the energy transfer relationships reflected in the spectral peak trajectories. In the steady-state re-entry segment, based on the target spectral pattern sequence that has been determined to meet the external steady-state conditions, a steady-state spectral benchmark is constructed for each installation location identifier. The steady-state spectral benchmark is used to characterize the node distribution characteristics, edge density characteristics, peak drift characteristics, and spike density characteristics of the corresponding installation location identifier under normal operating conditions. In the recovery and release segment, attractor regression is performed on the spectral pattern sequence and the steady-state spectral benchmark to obtain the evaluation results. The evaluation results include abnormal edge residuals, node bias, drift non-convergence, and spectral spike residuals. Under the premise that the external steady-state conditions are met, if it is determined that any sub-result in the evaluation results deviates from the steady-state spectral benchmark by a preset period, the corresponding recovery release segment is marked as a state of incomplete implicit recovery. In subsequent steady-state re-entry segments, the secondary anomaly triggered by load mutation is bound to the state of incomplete implicit recovery as a delayed recurrence homologous link based on the spectral inheritance relationship, so as to realize the identification and attribution of implicit recovery failure caused by the incomplete release of slow variables in the control loop.

[0007] A second aspect of this application provides a status monitoring device for a subway flexible DC power supply equipment of an energy router. The energy router's DC-side filter inductor, AC-side parallel reactor, busbar film capacitor, and power module heat dissipation base are each pre-configured with a vibration pickup unit. Each vibration pickup unit is bound to a unique installation location identifier and a time synchronization identifier. The device includes an acquisition module and a processing module. The acquisition module is used to acquire, during the operation of the energy router, the vibration pickup sequence sent by the vibration pickup unit and the operating status sequence bound to the energy router's operation process, and to compare the vibration pickup sequence with the operating status sequence based on the time synchronization identifier. The running state sequence is time-aligned to form an aligned sequence, which includes a fault-crossing flag, a current-limiting flag, and a recovery flag. The processing module is used to segment the aligned sequence based on the fault-crossing flag, the current-limiting flag, and the recovery flag to obtain target time segments, which include a drop-in segment, a current-limiting maintenance segment, a recovery-release segment, and a steady-state re-entry segment. The processing module is also used to perform structural acoustic pattern generation processing on the target time segments to obtain pattern segments, and to construct a pattern map based on the pattern segments to form a pattern map sequence. The nodes of the spectral pattern are represented by peak clusters in the spectral peak set, and the edges of the spectral pattern are represented by the energy transfer relationships reflected in the peak trajectories. The processing module is further configured to, within the steady-state reentry segment, construct a steady-state spectral reference for each installation location identifier based on the target spectral pattern sequence that has been determined to meet external steady-state conditions. The steady-state spectral reference is used to characterize the node distribution characteristics, edge density characteristics, peak drift characteristics, and spike density characteristics of the corresponding installation location identifier under normal operating conditions. The processing module is also configured to, within the recovery and release segment, perform attractor regression analysis on the spectral pattern sequence and the steady-state spectral reference. The evaluation results are obtained, including abnormal edge residuals, node bias, drift non-convergence, and spectral spike residuals. The processing module is further configured to, under the premise that the external steady-state conditions are met, if it is determined that any sub-result in the evaluation results deviates from the steady-state spectral reference by a preset period, mark the corresponding recovery release segment as a hidden recovery incomplete state, and bind the secondary anomaly triggered by the load mutation to the hidden recovery incomplete state as a delayed recurrence homologous link in the subsequent steady-state re-entry segment based on the spectral inheritance relationship, so as to realize the identification and attribution of the hidden recovery failure caused by the incomplete release of the slow variable in the control loop.

[0008] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, and both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described above.

[0009] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing instructions that, when executed, perform the method described above.

[0010] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By aligning the vibration pickup sequence with the operating state sequence in time, and introducing fault crossing, current limiting, and recovery flags for segmentation, the different control semantic stages of the energy router during complex operation are explicitly distinguished. This avoids mixing fundamentally different operating stages such as dropout, current limiting, recovery, and steady state for analysis, reducing semantic ambiguity in state determination from the source and establishing a clear and traceable temporal semantic foundation for subsequent feature extraction and state determination. By performing structural acoustic spectral generation processing on the target time segment and further constructing spectral patterns and spectral pattern sequences, the influence of slow variables in the control loop, which was originally difficult to observe directly, is transformed into stable and perceptible structural acoustic evolution features. This allows implicit changes in the internal control state that are "insignificant" to external manifestations to be indirectly revealed through structural features such as node distribution, edge relationships, spectral peak drift, and spectral spike behavior, significantly improving the sensitivity of state monitoring to complex control behaviors.

[0011] By constructing a steady-state spectral benchmark in the steady-state reentry segment, subsequent judgments no longer rely on static thresholds or empirical rules, but are based on a comparison with the structural acoustic benchmark formed by the equipment itself under normal operating conditions. This effectively adapts to the differences in installation locations, individual equipment, and operating environments, enhancing the self-consistency and generalization ability of monitoring results. Introducing attractor regression judgment in the recovery and release segment elevates state analysis from single-moment comparison to a judgment of the time evolution process. This enables the monitoring system to identify the dynamic behavior of "whether it is returning to steady state," rather than just "whether it has exceeded the limit," thus effectively distinguishing between short-term disturbances and structural anomalies caused by incomplete release of slow variables in the control loop, reducing the risk of false alarms and missed alarms. Through preset periodic constraints and spectral inheritance relationships, the implicit incomplete recovery state in the recovery phase is causally bound to the secondary anomalies triggered by subsequent load mutations at the structural acoustic level. This unifies the originally temporally separate and superficially independent anomaly events into the same delayed recurrence homogeneous link, solving the problem that existing monitoring systems struggle to identify the homogeneity of anomalies. This provides attribution results that are more consistent with actual operating mechanisms for fault analysis and maintenance decisions. Therefore, it facilitates the identification and attribution of implicit recovery failures caused by the incomplete release of slow variables in the control loop, thereby improving the accuracy of status monitoring of the power router's subway flexible DC power supply equipment. Attached Figure Description

[0012] Figure 1A flowchart illustrating a method for monitoring the status of a subway flexible DC power supply device using an energy router, as provided in an embodiment of this application; Figure 2 A schematic diagram of a status monitoring device for a subway flexible DC power supply equipment of an energy router provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0013] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0015] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0016] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0017] To address the aforementioned technical problems, this application provides a method for monitoring the status of subway flexible DC power supply equipment using an energy router, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a method for monitoring the status of a subway flexible DC power supply device using an energy router, as provided in an embodiment of this application. The method is applied to a server, and the DC-side filter inductor, AC-side parallel reactor, bus thin-film capacitor, and power module heat dissipation base of the energy router are pre-configured with vibration pickup units. Each vibration pickup unit is bound to a unique installation location identifier and a time synchronization identifier. The method includes steps S110 to S160, as follows:

[0018] S110. During the operation of the energy router, the vibration pickup sequence sent by the vibration pickup unit and the operation status sequence bound to the operation process of the energy router are acquired, and the vibration pickup sequence and the operation status sequence are time-aligned based on the time synchronization identifier to form an aligned sequence. The operation status sequence includes fault crossing flag, current limiting flag and recovery flag.

[0019] Specifically, a server is a computing device used for centralized processing, storage, and distribution of data and computational tasks. Its core characteristic is its ability to provide stable, controllable, and scalable services to multiple clients or system modules simultaneously under continuous operation. In energy router-related status monitoring scenarios, the server does not directly participate in power conversion or control execution, but rather exists as a computing carrier for information aggregation and analysis, focusing on data processing, status analysis, and result output. In the embodiments of this application, the server can be understood as a centralized or distributed computing platform that carries the status monitoring and analysis logic of the energy router.

[0020] During the operation of the energy router, the time synchronization identifier and installation location identifier are first fixed for each vibration pickup unit, and the generation and update rules of the time synchronization identifier are kept consistent in the vibration pickup sequence link and the operation status sequence link, so that the two links use the same timing reference and the same synchronization event source. A vibration pickup unit refers to a vibration or acoustic signal acquisition device located on components such as DC-side filter inductors, AC-side parallel reactors, busbar thin-film capacitors, or power module heat sinks. It outputs a pickup sequence related to structural micro-vibrations. The pickup sequence is a set of time-series signals continuously output by the pickup unit according to a sampling period; each pickup sample in the sequence corresponds to a specific time position. An installation location identifier is an identifier used to uniquely identify the physical installation point of the pickup unit, ensuring that the same installation point is always traceable and not confused with other installation points during subsequent processing. A time synchronization identifier is an identifier used to establish the same time reference between different signal links; its essence is a comparable timestamp or synchronization count value. A timing reference refers to a unified time reference for the entire system, such as the same hardware clock domain or the same control cycle counter. A synchronization event source refers to a unified event that triggers the alignment of the time synchronization identifier, such as the control cycle trigger edge, the sampling timer trigger edge, or the phase switching edge of the fault-crossing state machine. By selecting the same synchronization event source, the pickup sequence and the operating state sequence can be consistently identified and aligned on the time axis.

[0021] When acquiring the operating status sequence, the DC bus voltage, DC bus current, AC phase current, duty cycle sequence, fault ride-through flag, current limiting flag and recovery flag are synchronously sampled based on the synchronous event source. A time synchronization identifier consistent with the vibration pickup sample is written for each operating status sampling point, thereby ensuring that the operating status sequence and the vibration pickup sequence can be matched one-to-one according to the same time synchronization identifier. The operating state sequence refers to the multivariate time-series data set generated by the energy router during operation. Each sampling point contains electrical quantities and control state quantities at the same moment. The DC bus voltage refers to the instantaneous voltage measurement value at both ends of the DC bus of the energy router. The DC bus current refers to the instantaneous current measurement value in the DC bus circuit. The AC phase current refers to the instantaneous current measurement value of each phase of the AC interface. The duty cycle sequence refers to the duty cycle time sequence generated by the modulation process within a continuous control cycle, used to characterize the change of the duty cycle of the switching drive for power conversion. The fault ride-through flag refers to the state flag indicating whether the energy router is in fault ride-through operation semantics. The current limiting flag refers to the state flag indicating whether the energy router is in current limiting constraint semantics. The recovery flag refers to the state flag indicating whether the energy router is in fault recovery release semantics. Synchronous sampling refers to sampling the above electrical quantities and flag quantities at the same time point and writing them to the same time synchronization identifier under the trigger of the same synchronization event source. This avoids time misalignment caused by sampling electrical quantities first and flag quantities later, thus providing a consistent temporal semantic basis for subsequent segmentation and spectral modeling.

[0022] On the data aggregation side, an alignment buffer is established based on the time synchronization identifier. Dual-stream sorting and association are performed on the vibration pickup sequence and the running state sequence. When the vibration pickup sample and the running state sampling point corresponding to the same time synchronization identifier are detected, alignment entries are generated to form an alignment sequence, thereby coupling and solidifying the structural acoustic information and the running semantic information on the same time axis. The data aggregation side refers to the processing entity responsible for receiving, caching, sorting, and splicing multi-channel sampled data; in this embodiment, it is a server. The alignment buffer refers to a cache structure used to temporarily store the vibration sequence and the running status sequence and wait for pairing; its core index key is the time synchronization identifier. Dual-stream sorting refers to rearranging the vibration sequence and the running status sequence according to the time synchronization identifier to eliminate order disturbances caused by communication jitter, buffer delay, or packet out-of-order. Dual-stream association refers to pairing vibration samples and running status sampling points under the same time synchronization identifier with the time synchronization identifier as the matching condition. The alignment entry refers to the data record formed after a successful pairing, which includes at least the time synchronization identifier, the installation location identifier, the vibration sample or the set of vibration samples, and the corresponding running status sampling point. The alignment sequence refers to a sequence composed of multiple alignment entries in ascending order of the time synchronization identifier, enabling subsequent processing to directly perform segment division based on fault crossing flags, current limiting flags, and recovery flags on the alignment sequence, and to perform consistent structural acoustic pattern generation and pattern map construction on the vibration samples within each segment, avoiding misjudgment caused by inconsistencies between the pattern map sequence and the semantics of the running stage due to cross-link time misalignment.

[0023] S120. Based on the fault crossing flag, current limiting flag, and recovery flag, the alignment sequence is segmented to obtain the target time segment. The target time segment includes the drop-in segment, the current limiting maintenance segment, the recovery release segment, and the steady-state re-entry segment.

[0024] Specifically, when segmenting the alignment sequence based on the fault crossing flag, the rate limiting flag, and the recovery flag, each alignment entry in the alignment sequence is traversed first, and the time synchronization flag and the fault crossing flag, the rate limiting flag, and the recovery flag bound to the time synchronization flag are read from each alignment entry. This makes the time synchronization flag form a monotonically increasing index sequence on the time axis, and at the same time, the fault crossing flag, the rate limiting flag, and the recovery flag form a flag sequence that corresponds one-to-one with the alignment sequence on the time synchronization flag dimension. Alignment entries refer to record units formed by pairing the vibration sample set with the operating state sampling points using the same time synchronization identifier as the primary key; time synchronization identifier refers to a unified time index used to align the vibration sequence and the operating state sequence, which can be represented as a synchronization count value or a unified timestamp; flag sequence refers to a discrete state sequence arranged in the order of time synchronization identifiers, where each position corresponds to the flag value of a sampling time; fault crossing flag indicates whether the energy router is in the fault crossing semantic stage, current limiting flag indicates whether the energy router is in the current limiting constraint semantic stage, and recovery flag indicates whether the energy router is in the recovery release semantic stage. By forming a flag sequence, the positioning of all subsequent segment boundaries is constrained within the unified coordinate system of the time synchronization identifier, thereby ensuring that the segment division and the vibration sample set remain consistent in time.

[0025] When performing flag debouncing on the flag sequence, a state stability judgment window is established for the fault crossing flag, current limiting flag, and recovery flag. At each time synchronization marker position, it is determined whether the flag value remains consistent within the continuous stability judgment window. Only when the flag value remains consistently consistent within the stability judgment window is the flag state confirmed as valid, thereby eliminating instantaneous false flips caused by communication jitter, sampling jitter, or control signal glitches. Flag debouncing refers to the process of stabilizing a discrete state sequence. The stability judgment window refers to the continuous time synchronization marker interval used to confirm the persistence of the state. Instantaneous false flips refer to non-real state changes that occur in a very short time. If the state confirmation rules for flag debouncing are expressed in functional form, the original flag sequence can be denoted as the original flag sequence function, and the debouncing flag sequence as the debouncing flag sequence function. The stability judgment window length is denoted as the window length. Then, the debouncing flag value at the time synchronization marker index position can be determined by majority voting within the window, thus obtaining a debouncing flag sequence that is insensitive to glitches. The specific formula is as follows:

[0026]

[0027] in, Indicates the original flag sequence at index position The value at this location is either 0 or 1. Indicates the dejitter flag sequence at index position The value at this location is either 0 or 1. This represents the window length and the number of consecutive-time synchronization flags participating in the stability assessment. Take a positive integer and can be set to greater than or equal to 3 to suppress single-point spikes; This represents the majority voting function, whose output is the value that appears most frequently in the set. By using majority voting, the value of the de-jitter flag will not be changed by a single or a few erroneous flips, thereby improving the stability of segment boundary recognition.

[0028] After completing the flag debouncing process, when identifying fault crossing entry events, rate limiting entry events, recovery entry events, and recovery exit events, the rising and falling edges of the debouncing flag sequence are detected respectively. The position of the time synchronization flag that first meets the rising or falling edge condition is recorded as the position index of the corresponding event. An event refers to the critical moment defined by the flag state flip. The fault crossing entry event is the moment when the fault crossing flag flips from 0 to 1. The rate limiting entry event is the moment when the rate limiting flag flips from 0 to 1. The recovery entry event is the moment when the recovery flag flips from 0 to 1. The recovery exit event is the moment when the recovery flag flips from 1 to 0. The rising edge refers to the flip of the flag from 0 to 1, and the falling edge refers to the flip of the flag from 1 to 0. Through event recognition, the continuous operation process can be anchored into several time points with clear semantic boundaries, providing directly referable boundary indexes for subsequent continuous interval division.

[0029] Based on the positions of fault crossing entry events, flow limiting entry events, recovery entry events, and recovery exit events on the time synchronization identifier, when dividing the alignment sequence into continuous intervals, the event position index is used as the dividing boundary. The continuous intervals on the time synchronization identifier axis are mapped to drop entry segments, flow limiting maintenance segments, recovery release segments, and steady-state re-entry segments, and each segment corresponds to a continuous alignment entry subsequence in the alignment sequence. The continuous interval division refers to selecting an uninterrupted index interval on the time synchronization marker axis. The drop-in segment refers to the continuous interval from the start of the fault crossing entry event to the end before the current limiting entry event, used to characterize the transition phase after voltage drop triggering and before entering current limiting. The current limiting maintenance segment refers to the continuous interval from the start of the current limiting entry event to the end before the recovery entry event, used to characterize the maintenance phase of the continuous triggering of the current limiting flag. The recovery release segment refers to the continuous interval from the start of the recovery entry event to the end of the recovery exit event, used to characterize the release phase of the continuous triggering of the recovery flag. The steady-state re-entry segment refers to the continuous interval from the start of the recovery exit event to the end before the next fault crossing entry event or to the end of the preset steady-state observation window, used to characterize the steady-state regression phase after recovery exit. The above mapping relationship ensures that each segment is continuous in time and has a single operational semantic, thereby avoiding distortion of spectral pattern modeling and attractor regression judgment caused by cross-semantic mixing.

[0030] After segmentation, during segment consistency verification for each target time segment, the dejittering flag sequence within the time synchronization identifier range covered by the target time segment is checked one by one to see if it meets the preset operational semantic constraints. Target time segments that do not meet the preset operational semantic constraints are marked as boundary untrusted segments to trigger boundary correction processing. Segment consistency verification refers to the verification process of the persistence relationship of flag states within a segment. Preset operational semantic constraints refer to the set of constraints on the flag combination relationships that each segment should satisfy. Boundary untrusted segments refer to segments whose segment start time synchronization identifier or segment end time synchronization identifier may be incorrectly located by glitches or jitter. Boundary correction processing keeps the time synchronization identifier as the unique time coordinate unchanged and introduces lookback and lookforward windows near the boundary to reposition the first stable rising edge and the first stable falling edge, so that the segment start time synchronization identifier and segment end time synchronization identifier are realigned to a more stable event position index. This ensures that the corrected target time segment is consistent with the aligned sequence in temporal semantics and provides a stable and repeatable segment boundary for subsequent structural acoustic pattern generation processing.

[0031] S130. Perform structural acoustic pattern generation processing on the target time segment to obtain pattern fragments, and construct pattern maps based on the pattern fragments to form a pattern map sequence. The nodes of the pattern map are represented by the peak clusters in the peak set, and the edges of the pattern map are represented by the energy transfer relationships reflected by the peak trajectories.

[0032] Specifically, when performing preprocessing on the pickup sequence within the target time segment, firstly, a continuous set of pickup samples is extracted from the alignment sequence corresponding to the target time segment according to the installation location identifier and spliced ​​into a pickup sequence, ensuring that the pickup sequence remains monotonically increasing on the time synchronization identifier and does not cross the segment boundary of the target time segment; DC bias removal is used to eliminate baseline offsets caused by installation stress of the pickup unit, sensor zero-point drift, etc., so that the subsequent spectral structure is not masked by low-frequency bias; amplitude normalization is used to eliminate the coupling strength difference between different installation location identifiers, so that the energy distribution of subsequent spectral peaks is comparable; bandwidth limitation is used to limit the pickup sequence to the frequency band where the pickup unit has a reliable response and the structural acoustic information is concentrated; pulse artifact suppression is used to remove isolated spikes caused by communication jitter, poor contact, and transient electromagnetic interference, preventing spikes from expanding into broadband pseudo-spectral spikes in the frequency domain and interfering with the extraction of spectral peak sets. The preprocessing can be obtained by the following combined form:

[0033] in, Indicates the sampling sequence number Vibration pickup samples at the location; This represents the sequence after DC bias removal is complete; This represents the length of the DC bias estimation window, which is a positive integer and covers multiple sampling points. The weighting function for DC bias estimation is used to increase the contribution of neighboring samples to the bias estimation. The weighting function takes a non-negative value and can be in a decreasing form. The above expression estimates the local baseline by weighted moving average and subtracts it from the original sequence, so that slow drift is suppressed while the relative change of structural vibration is preserved.

[0034]

[0035] in, This represents the sequence after amplitude normalization; This represents the reference mean, which can be derived from the target time segment. Statistical results or baseline statistics from steady-state reentry segments; Represents the reference variance, which can be compared with... Use consistent statistical intervals; The term represents the stability term, used to avoid instability in the denominator due to excessively small reference variance. It is a positive number and a small value. The above expression compresses the influence of amplitude scales of different channels on the energy distribution of spectral peaks by unifying the mean and fluctuation scale.

[0036]

[0037] in, This represents a sequence after bandwidth limitation; This represents the impulse response of the bandpass filter. The lower and upper passband frequencies of the filter are preset based on the frequency response of the pickup unit and the structural modal distribution. The filter order is represented by a positive integer and determines the steepness of the filter transition band. The above expression achieves bandpass filtering through convolution, which suppresses low-frequency drift and high-frequency noise outside the target frequency band.

[0038]

[0039] in, This indicates the preprocessed pickup sequence; Indicates The neighborhood sampling set centered on the center, the neighborhood length is determined by a preset neighborhood window; Represents the median function; This represents the median absolute deviation function, used to characterize the robustness level of fluctuations within the neighborhood; This represents the impulse discrimination coefficient. A positive number and a larger value indicate that it is less likely to classify a sample as an impulse. This represents the suppression intensity coefficient, which is a number between 0 and 1. The larger the value, the stronger the reduction in pulse deviation; The function is an indicator function that takes the value 1 when the condition is met and 0 otherwise. The above expression identifies isolated spikes through robust neighborhood statistics and reduces their deviation by suppressing the intensity, thereby reducing the interference of pulse artifacts on the spectral spike density in the frequency domain.

[0040] When performing short-time window segmentation to form spectral window segments based on the preprocessed pickup sequence, the corresponding sampling sequence pairs are identified along the time synchronization. A sliding truncation is performed, and a spectral window time synchronization identifier is bound to each truncation interval, ensuring a one-to-one correspondence between the spectral window time synchronization identifier and the time synchronization identifier interval covered by the spectral window segment. The purpose of short time window segmentation is to convert the non-stationary pickup sequence into a series of approximately stationary spectral window segments, enabling the extraction of spectral peak sets and the tracking of spectral peak trajectories to establish a comparable frequency domain structure between continuous spectral windows. The spectral window segment can be represented as:

[0041] in, The spectral window time synchronization flag is indicated as Spectral window fragments; This represents the spectral window index, which increments sequentially by time. It represents the frame shift length, takes a positive integer, and determines the degree of overlap between adjacent spectral window segments; This represents the sample index within the spectral window segment, with a value ranging from 0 to... ; The length of the spectral window is represented by a positive integer and covers multiple sampling points that cover at least one major period of structural vibration. The window function is used to reduce spectral leakage and highlight the stable structure in the central region of the spectral window. The window function takes a non-negative value and approaches 0 at the boundary. The above expression uses sliding truncation and window function weighting to make the spectral estimation more stable in locality, while controlling the trade-off between time resolution and frequency resolution by the frame shift length.

[0042] When performing peak set extraction for each spectral window segment and peak trajectory tracking between adjacent spectral window segments, the spectral window spectrum is first calculated for the spectral window segment to obtain the peak set and peak energy distribution. Then, the peaks are correlated between adjacent spectral windows based on frequency proximity, energy continuity, and bandwidth continuity to form peak trajectories. The drift direction, drift rate, and spike density are extracted from the peak trajectories. The spectral window spectrum can take the following form:

[0043] in, The spectral window time synchronization flag is indicated as The complex spectrum; This represents the angular frequency variable, whose range of values ​​is determined by the discrete frequency sampling set; The imaginary unit is represented; the above expression maps a time-domain spectral window segment to the frequency domain, causing the frequency components corresponding to structural resonances to appear as spectral peaks, the energy distribution of which can be derived from... Distribution characterization on the frequency axis; spectral peak set extraction through... Local maxima detection is performed on the peaks, and minimum peak height and minimum peak spacing constraints are applied. Each spectral peak contains the peak center frequency, peak amplitude, and peak bandwidth. Peak trajectory tracking can be achieved through minimum cost association, connecting the peaks of the same structural mode in adjacent spectral windows into a continuous trajectory.

[0044] in, Indicates that the spectral window time synchronization is marked as The index of the selected spectral peak. This represents the spectral peak index sequence across the spectral window. Indicates the optimal spectral peak trajectory; Indicates the number of spectral window segments; This represents the center frequency of the spectral peak, and its value is given by the spectral peak detection results. This represents the amplitude of the spectral peak, and its value is given by the spectral peak detection results. This represents the bandwidth of the spectral peak, and its value is given by the spectral peak detection results. , , The cost weight is represented by a non-negative number and used to balance frequency continuity, amplitude continuity, and bandwidth continuity. This expression minimizes the cost of change between adjacent spectral windows, suppressing erroneous jumps caused by noise, thus making the spectral peak trajectory more consistent with the slow time-varying characteristics of physical modes. The drift direction is characterized by the increasing or decreasing trend of the spectral peak center frequency in the time dimension, and the drift rate is characterized by the magnitude of the change in the spectral peak center frequency relative to the spectral window time synchronization marker. The spectral spike density is used to measure the sharp energy spikes formed by broadband transient disturbances in the frequency domain, and can be obtained by statistically analyzing the proportion of frequency points exceeding the robust background threshold.

[0045] in, The spectral window time synchronization flag is indicated as spectral spike density; Represents the number of discrete frequency points, and is a positive integer. Indicates the first A discrete frequency point; express The set at all discrete frequency points; This represents the spectral spike threshold coefficient; a positive number indicates a more stringent threshold. and Used to construct a robust background level so that broadband transient spikes can be stably identified; the above expression makes the spike density reflect the frequency domain occupancy of transient disturbances by robustly thresholding the spectral energy and counting the proportion of points exceeding the threshold.

[0046] When constructing and encoding spectral peak clusters for each spectral window segment based on spectral peak sets, the spectral peak sets are aggregated according to frequency proximity, bandwidth overlap, and energy distribution similarity to obtain spectral peak clusters. These clusters are used to represent the stable structure of the same resonance or harmonic cluster, rather than treating interconnected peaks as isolated peaks. Spectral peak clusters can be constructed through spectral clustering of similarity maps to ensure stable clusters can still be formed even when spectral peaks are dense and harmonic structures exist.

[0047] in, Indicates the first The first spectral peak and the first The similarity of the spectral peaks; , Indicates the center frequency of the spectral peak; , Indicates the amplitude of the spectral peak; , Indicates the bandwidth of the spectral peak; It represents a measure of bandwidth overlap, which can be calculated from the overlap ratio of the bandwidth intervals of two spectral peaks; , , The similarity scale parameter is positive and determines the sensitivity of frequency difference, amplitude difference, and bandwidth overlap to similarity. The above expression assigns higher similarity to spectral peaks with similar frequencies, similar amplitudes, and significant bandwidth overlap, making them more likely to be clustered into the same spectral peak cluster. After obtaining the spectral peak cluster, the spectral window time synchronization identifier, spectral peak cluster, spectral peak trajectory, drift direction, drift rate, and spectral spike density are jointly encoded into spectral pattern fragments, making the spectral pattern fragments the sole input carrier for subsequent mapping. Among them, the spectral window time synchronization identifier is used to align the temporal sequence, the spectral peak cluster is used to express node candidates, the spectral peak trajectory is used to express cross-spectral window connection clues, the drift direction and drift rate are used to express dynamic evolution, and the spectral spike density is used to express transient perturbation background.

[0048] When constructing a spectral pattern map and forming a spectral pattern map sequence based on spectral pattern fragments, a spectral pattern map is constructed using the spectral pattern fragment corresponding to each spectral window time synchronization identifier. These fragments are then concatenated in ascending order of the spectral window time synchronization identifiers to obtain the spectral pattern map sequence, ensuring consistency between the spectral pattern map sequence and the target time segment type identifier. Nodes in the spectral pattern map are represented by spectral peak clusters, each carrying node attributes such as cluster center frequency, cluster energy distribution, and cluster stability. These node attributes are inherited from the statistical aggregation results of the spectral peak clusters. Edges in the spectral pattern map are represented by energy transfer relationships, determined by the inter-cluster migration of spectral peak trajectories between adjacent spectral windows. Edge weights characterize the migration intensity, thus solidifying the dynamic process of how spectral peaks transition from one cluster to another into edges in the graph structure. Edge weights can take the following forms:

[0049] in, The spectral window time synchronization flag is indicated as Time from node To the node Edge weights; Indicated in the spectral window With Spectral Window The set of spectral peak trajectories tracked between them; Indicates a spectral peak trajectory identifier; Representing the trajectory In the spectral window The value is the cluster number of the spectral peaks to which it belongs, and is taken as the node index; Representing the trajectory In the Spectrum Window The spectral cluster number to which it belongs; Representing the trajectory In the spectral window The amplitude or energy representation value of the corresponding spectral peak; The stable term is used to avoid instability in normalization due to an excessively small denominator. The above expression normalizes the inter-cluster migration of statistical trajectories between adjacent spectral windows according to energy proportion, so that the edge weights can reflect which nodes the main energy flows from to which nodes, thus providing structured input for subsequent node distribution characteristics, edge density characteristics, and attractor regression determination. When splicing the spectral pattern sequence, each spectral pattern in the sequence is bound to the same target time segment type identifier, and splicing across target time segments is prohibited. This ensures that the spectral pattern sequences of the fall-in segment, current-limited maintenance segment, recovery release segment, and steady-state re-entry segment are semantically singular and can be directly used for subsequent steady-state spectral benchmark construction and evaluation result generation.

[0050] S140. In the steady-state reentry segment, based on the target spectral pattern sequence that has been determined to meet the external steady-state conditions, a steady-state spectral reference is constructed for each installation location identifier. The steady-state spectral reference is used to characterize the node distribution characteristics, edge density characteristics, peak drift characteristics, and spike density characteristics of the corresponding installation location identifier under normal operating conditions.

[0051] Specifically, when determining candidate intervals in the steady-state reentry segment, each alignment entry is first traversed in the alignment sequence according to the time synchronization identifier. Then, the running state sequence sampling points corresponding one-to-one with the time synchronization identifier are read from each alignment entry, forming a traceable state timeline in the time synchronization identifier dimension. Subsequently, the values ​​of the fault crossing flag, current limiting flag, and recovery flag are simultaneously checked on this state timeline, requiring all three to be in an untriggered state. The continuous time synchronization identifier interval that meets this requirement is defined as the candidate interval. The candidate interval refers to the continuous time range within which fault crossing, current limiting, and recovery state remnants have been excluded at the operational semantic level. The flag consistency constraint refers to the constraint condition on the flag combination relationship within the candidate interval. Its purpose is to ensure that the subsequent spectral pattern sequence used to construct the steady-state spectral pattern benchmark originates from explicit steady-state reentry semantics, rather than from transitional semantics of fault crossing remnants or incomplete recovery release, thereby avoiding contamination of the steady-state spectral pattern benchmark by non-steady-state structural acoustic features.

[0052] When determining external steady-state conditions and establishing a set of steady-state intervals within candidate intervals, stability metrics are constructed for DC bus voltage, DC bus current, and AC phase current, respectively. The set of steady-state intervals is then compiled from continuous-time synchronization intervals that simultaneously satisfy preset conditions. External steady-state conditions refer to observable stable operating conditions at the level of external electrical quantities, ensuring the premise that the external environment is already in a steady state. The set of steady-state intervals refers to a collection of multiple steady-state intervals, each corresponding to a continuous-time synchronization range and satisfying the external steady-state conditions. Stability metrics can employ a joint constraint of rolling fluctuation intensity and trend drift intensity to avoid missing slow drifts due to relying solely on variance thresholds.

[0053]

[0054] in, Indicates the signal Time synchronization identifier index position The stability index obtained from the calculation. Any one of the DC bus voltage, DC bus current, or AC phase current can be selected; This represents the length of the scrolling window, is a positive integer, and is used to limit the time range for stability evaluation. This represents the mean value within the scrolling window; This represents the weighting coefficients for the fluctuation and drift terms, with values ​​ranging from 0 to 1. A larger value indicates a greater emphasis on fluctuation intensity constraints; The stability term, represented by a positive number, is used to avoid numerical instability due to an excessively small denominator. The first term normalizes the relative fluctuations within the rolling window, while the second term characterizes the trend drift intensity by the ratio of net change to total change. When the signal exhibits only small random fluctuations without sustained drift within the window, the stability index tends to be small. The stability indices calculated for DC bus voltage, DC bus current, and AC phase current are simultaneously compared with their respective thresholds, and the threshold conditions must be met across several consecutive time synchronization markers to form a set of steady-state intervals. This provides a strict time boundary for subsequent spectral pattern sequence trimming.

[0055] When performing interval pruning on the spectral pattern sequence within the steady-state reentry segment based on the steady-state interval set, each spectral pattern in the sequence is first mapped to its bound spectral window time synchronization marker range, and the spectral window time synchronization marker range is projected onto the time synchronization marker axis, so that the spectral pattern sequence can be aligned with the steady-state interval set in the same time synchronization marker coordinate system. Subsequently, for each steady-state interval, the spectral pattern subsequence that completely falls within the steady-state interval's spectral window time synchronization marker range is truncated, and all spectral pattern subsequences are merged into the target spectral pattern sequence. Interval pruning refers to the selective truncating of the spectral pattern sequence according to time boundaries; the target spectral pattern sequence refers to the sequence that is strictly time-aligned with the steady-state interval set and contains only spectral patterns within the steady-state interval. Its purpose is to ensure that the sample source of the steady-state spectral pattern benchmark is synchronously established with the external steady-state conditions, and to avoid introducing bias by mixing in spectral patterns that are not yet fully stable near the steady-state determination boundary.

[0056] When extracting a set of spectral patterns from the target spectral pattern sequence for each installation location identifier and performing node alignment and edge alignment processing, the target spectral pattern sequence is first grouped according to the installation location identifier to obtain the set of spectral patterns corresponding to that installation location identifier; then, cross-... Figure 1 Consistent node and edge categories ensure that peak cluster nodes under different spectral window time synchronization identifiers can be merged into the same steady-state node category, and energy transfer edges under different spectral window time synchronization identifiers can be merged into the same steady-state edge category. Node alignment refers to the process of mapping nodes in different spectral patterns to a unified node category space, avoiding the misidentification of the same physical mode as different nodes due to small drifts in the center frequency of peak clusters; edge alignment refers to the process of mapping edges in different spectral patterns to a unified edge category space, avoiding the misidentification of the same connection mode as different edges due to short-term fluctuations in energy transfer intensity. Node alignment can be performed across graphs using a joint distance of frequency proximity and similar energy distribution.

[0057]

[0058] in, Indicates that the node With nodes Distance when considered as a candidate node of the same type; , This represents the cluster center frequency of the spectral peak cluster corresponding to the node, and its value is given by the result of the spectral peak cluster construction. , This indicates that the spectral peak cluster corresponding to the node is in the preset frequency bin. Cluster energy distribution components on Indicates the number of boxes, rounded to a positive integer; The first term represents the weighting coefficients of the frequency term and the energy distribution term, with values ​​ranging from 0 to 1; the second term uses the overlap ratio of the energy distribution to characterize the similarity, with a larger overlap indicating a smaller distance. For stability terms, a positive value is used; this suppresses erroneous splitting caused by frequency drift and erroneous merging caused by abrupt changes in energy distribution. Based on this, nodes with a distance less than a preset threshold can be merged into the same steady-state node category. Edge alignment is performed in the node category space. The endpoints of the edges in the spectral pattern are first mapped to steady-state node categories, and then the endpoint category pairs are used as steady-state edge category identifiers, thus forming a set of steady-state node categories and a set of steady-state edge categories.

[0059] Node distribution characteristics are extracted based on the steady-state node category set, and edge density characteristics are extracted based on the steady-state edge category set. Simultaneously, peak drift and spike density characteristics are extracted. First, the frequency of occurrence, energy proportion, and temporal stability of the spectral pattern set at the node category level are statistically analyzed to form node distribution characteristics. Then, the frequency of occurrence, weight proportion, and persistence of edges at the edge category level are statistically analyzed to form edge density characteristics. Simultaneously, the consistency of drift direction and the range of drift rate are statistically analyzed for the spectral peak trajectory set to form peak drift characteristics. Finally, the background level and fluctuation range of the spike density sequence within the spectral window are statistically analyzed to form spike density characteristics. Node distribution characteristics refer to the resident degree and energy structure of each steady-state node category under steady-state conditions. Edge density characteristics refer to the density and strength structure of energy transfer relationships under steady-state conditions. Peak drift characteristics refer to the convergence and repeatability of the center frequency of the spectral peak cluster over time under steady-state conditions. Spike density characteristics refer to the background occupancy level of transient broadband spikes in the frequency domain under steady-state conditions. Node distribution characteristics can be expressed using a combination of frequency vector and energy vector, for example, for steady-state node categories. Its frequency and energy percentage are defined as follows:

[0060]

[0061]

[0062] in, Indicates the number of spectrograms in the spectrogram set; Indicates the first The set of nodes in the Zhangpuwen diagram; Indicates an indicator function; Indicates that the node A mapping function that maps to steady-state node categories; Represents a node In the The node energy representative value in the spectral pattern diagram can be obtained by aggregating the cluster energy distribution. Indicates the steady-state node category The frequency of occurrence in the steady-state spectral pattern set ranges from 0 to 1; Indicates the steady-state node category The energy percentage, with a value ranging from 0 to 1; For stability terms, the edge density characteristic can be used to statistically analyze the frequency and weight of steady-state edge categories to obtain a stable structural baseline; the spectral peak drift characteristic can characterize the quantile range of drift rate, and the spectral spike density characteristic can characterize the quantile range of spectral spike density, so that subsequent attractor regression judgment can be compared based on the steady-state allowable range rather than a single threshold.

[0063] After feature extraction, when performing steady-state pattern screening and reconstructing features based on the screened pattern set, the extracted node distribution characteristics, edge density characteristics, peak drift characteristics, and spike density characteristics are used as references. The deviation degree of each pattern in the target pattern sequence is calculated, and patterns with deviations exceeding a preset threshold are marked as anomalous and removed, thus forming the screened pattern set. Subsequently, the node distribution characteristics, edge density characteristics, peak drift characteristics, and spike density characteristics are recalculated only based on the screened pattern set, making the steady-state pattern benchmark closer to the steady-state dominant mode without being skewed by a small number of anomalous samples. Steady-state pattern screening refers to robust processing to further exclude occasional perturbations within the steady-state range. Anomalous patterns refer to patterns that, although within the steady-state range, have a graph structure or transient perturbation level that significantly deviates from the steady-state dominant mode. The deviation degree can be measured using a robust distance weighted by multiple features, for example, unifying node distribution deviation, edge density deviation, drift deviation, and spike deviation into a single screening index.

[0064]

[0065] in, Indicates the first The degree of deviation of the spectral pattern diagram; Represents the frequency component in the node distribution characteristics With energy components ; Indicates the first Zhang spectral pattern in steady-state node categories The corresponding component values ​​on, This represents the reference component value obtained from the statistics of the current set; This represents the median absolute deviation function, used to provide a robust metric that makes the screening insensitive to a small number of outliers; Indicates the first The drift rate represented by the spectral pattern is shown in the diagram. This represents a reference value for the drift rate; Indicates the first The density of spectral patterns corresponding to the pattern diagram. Indicates the reference value for spectral spike density; , , The weighting coefficient is a non-negative number used to balance the impact of various deviations on the screening results. Deviations in different characteristic dimensions are normalized according to a robust scale and weighted summed so that abnormal spectral patterns can be identified and removed when they deviate significantly in any dimension. After reconstruction, the set of steady-state node categories, the set of steady-state edge categories, and the reconstructed node distribution characteristics, edge density characteristics, spectral peak drift characteristics, and spectral spike density characteristics corresponding to each installation location are encapsulated into a steady-state spectral benchmark. The steady-state spectral benchmark is then bound to the installation location identifier for storage, enabling a one-to-one comparison with the installation location identifier during subsequent attractor regression judgment within the recovery and release segment. This ensures that the source range, screening logic, and statistical results of the steady-state spectral benchmark are traceable, reproducible, and auditable.

[0066] S150. Within the recovery and release fragment, attractor regression is performed on the spectral pattern sequence and the steady-state spectral pattern benchmark to obtain the evaluation results, which include the residual degree of abnormal connections, node bias, drift non-convergence, and spectral spike residual degree.

[0067] Specifically, when extracting and grouping the spectral pattern sequence corresponding to the recovery and release segment based on the time synchronization identifier, the start time synchronization identifier and end time synchronization identifier of the recovery and release segment are read first. Then, the spectral pattern of the spectral window time synchronization identifier falling into the time synchronization identifier interval after being projected onto the time synchronization identifier axis is filtered in the full spectral pattern sequence to form the spectral pattern sequence corresponding to the recovery and release segment. Subsequently, the installation position identifier of each spectral pattern in the spectral pattern sequence is read, and the installation position identifier is used as the grouping key to splice the spectral pattern corresponding to the same installation position identifier into a spectral pattern subsequence in ascending order of the spectral window time synchronization identifier, so that each spectral pattern subsequence corresponds one-to-one with the steady-state spectral reference corresponding to the installation position identifier. The recovery release segment refers to the target time segment defined from the recovery entry event to the recovery exit event, used to characterize the stage where the external output gradually recovers but the internal control variables may still be released; the spectral pattern sequence refers to the graph structure sequence ordered by the spectral window time synchronization identifier, and each spectral pattern consists of spectral peak cluster nodes and energy transfer edges; the spectral window time synchronization identifier refers to the time identifier used to index the spectral window segment, which can be aligned with the time synchronization identifier through a mapping relationship; the installation location identifier refers to the identifier used to uniquely identify the physical installation point of the vibration pickup unit, ensuring that the structural acoustic responses of different components are not mixed; the spectral pattern subsequence refers to the subsequence extracted from the spectral pattern sequence under the same installation location identifier constraint, the purpose of which is to enable attractor regression determination to compare the temporal evolution on the same spatial object.

[0068] When performing structural alignment processing for each installation location identifier, the steady-state spectral reference corresponding to that installation location identifier is first read, and the set of steady-state node categories and the set of steady-state edge categories in the steady-state spectral reference are used as a unified reference space. Then, for each spectral map in the spectral map subsequence, node mapping is first performed on the spectral peak clusters at the node level, so that the spectral peak clusters are mapped to steady-state node categories, deviating from steady-state node categories, or newly added abnormal node categories. Then, edge mapping is performed on the energy transfer relationships at the edge level, so that the energy transfer relationships are mapped to steady-state edge categories, missing steady-state edges, or newly added abnormal edges. In this way, the graph structure changes in the recovery and release segment are uniformly projected into the category space defined by the steady-state spectral reference. Structural alignment refers to converting the spectral patterns at different time points into comparable representations within the same coordinate system. Steady-state node categories refer to the node categories obtained through node alignment in the steady-state reentry segment, used to characterize the stable peak cluster structure under normal operating conditions. Deviation from steady-state node categories refers to situations where a peak cluster cannot find a matching object satisfying a distance threshold in the steady-state node category set but still exhibits partial similarity to the steady-state node category, used to characterize a slight shift. Newly added anomalous node categories refer to situations where a peak cluster cannot establish a reliable match with the steady-state node category set in terms of both frequency structure and energy distribution, used to characterize new anomalous resonance structures or new coupling modes. Steady-state edge categories refer to the edge categories obtained through edge alignment in the steady-state reentry segment, used to characterize energy transfer modes that are stable under normal operating conditions. Missing steady-state edges refer to situations where steady-state edge categories do not appear in the current spectral pattern, used to characterize missing connection modes. Newly added anomalous edges refer to energy transfer relationships that appear in the current spectral pattern but cannot be mapped to the steady-state edge category set, used to characterize new energy migration paths. To avoid mismatches caused by relying solely on frequency proximity in node mapping, a joint distance consisting of frequency distance, energy distribution distance, and local edge pattern distance can be used for node mapping.

[0069]

[0070] in, This indicates the nodes in the spectral pattern diagram. Mapping to steady-state node categories The joint distance; Represents a node The cluster center frequency corresponding to the spectral peak cluster, Indicates the steady-state node category The reference center frequency, and the values ​​of both are obtained from spectral peak cluster statistics and steady-state reference statistics; Represents a node In frequency division Cluster energy distribution components on Indicates the steady-state node category The reference energy distribution component, Indicates the number of frequency bins; Represents a node In the current spectral pattern map, the local edge pattern vector can be obtained by projecting the weights of adjacent edges of nodes onto the steady-state node category space. Indicates the steady-state node category Reference local connection pattern vector; Represents a norm function; , , Represents the weight coefficient, taking non-negative values ​​and satisfying the following conditions: In order to be interpreted as a weighted combination; For stable terms, the frequency position, energy distribution pattern, and local connectivity structure of nodes are constrained to reduce mismapping caused by relying on a single feature. Based on this, the stable node category with the smallest distance and less than the mapping threshold is used as the mapping result; otherwise, it is judged as a deviation from the stable node category or a newly added abnormal node category. Edge mapping is performed after node mapping is completed. After mapping the endpoints of the edges to the stable node category space, the edge category is determined by the endpoint category pair and the edge weight interval, thereby realizing the determination of stable edge categories, missing stable edges, and newly added abnormal edges.

[0071] After structural alignment, when performing attractor regression analysis along the spectral window time synchronization marker direction, the node distribution characteristics, edge density characteristics, spectral peak drift characteristics, and spectral spike density characteristics characterized by the steady-state spectral pattern reference are defined as the target attractor state. The structural alignment result at each spectral window time synchronization marker within the recovery-release segment is mapped to the current state vector, so that the evolution of the spectral pattern subsequence on the time axis is represented as a regression process of the current state vector towards the target attractor state. By tracking the regression process, it is determined whether there are cases where external steady-state conditions are met but the internal structural acoustic state has not yet regressed. Attractor regression analysis refers to treating the steady-state spectral pattern reference as the stable attractor state of the system and using a time series approach to evaluate whether the structural state within the recovery-release segment gradually approaches the attractor. The target attractor state refers to the reference state composed of the characteristic statistical results given by the steady-state spectral pattern reference. The regression process refers to the trajectory of the current state vector gradually approaching the target attractor state as the spectral window time synchronization marker advances. To achieve quantifiable tracking, the node distribution characteristics and edge density characteristics can be encoded as vectors, and a comprehensive regression distance can be defined.

[0072]

[0073] in, The spectral window time synchronization flag is indicated as The overall regression distance at the location; This represents the node distribution vector of the current spectral pattern in the steady-state node category space, and its components can be composed of the frequency of node occurrence or the proportion of node energy. The node distribution vector representing the steady-state spectral reference; This represents the edge density vector of the current spectral pattern in the steady-state edge category space. Its components can be composed of the frequency of edge occurrence or the weight ratio of edge. The edge density vector representing the steady-state spectral reference; This represents the drift rate value corresponding to the current spectral window. The drift rate reference value represents the steady-state spectral reference. This indicates the spectral spike density corresponding to the current spectral window. The reference value for the spectral spike density represents the steady-state spectral pattern reference. , , , This represents the weighting coefficient, which is a non-negative number and is used to balance the contribution of different characteristics to the regression distance; For stability terms, the differences between nodes and edges after structural alignment, as well as the differences between dynamic drift and transient disturbances, are unified into comparable distance quantities and observed on the time axis. Whether it shows a convergence trend, thus providing a dynamic basis for the calculation of subsequent evaluation results.

[0074] When calculating node bias, anomalous link remnant, drift nonconvergence, and spectral spike remnant based on attractor regression analysis results, quantifiable indicators are formed at the node, link, peak trajectory, and transient perturbation levels, respectively. These four indicators are then combined on the same spectral window time-synchronous marker axis to form the evaluation result. This allows the evaluation result to simultaneously cover four independent but complementary anomalous dimensions: structural distribution deviation, connection mode remnant, dynamic drift nonconvergence, and broadband perturbation remnant. Node bias refers to the degree of deviation of the current node distribution from the steady-state node distribution, used to characterize whether the peak cluster structure has returned to steady state. Anomalous link remnant refers to the degree to which newly added anomalous links and missing steady-state links persist within the recovery release segment, used to characterize whether the energy transfer relationship has completed annealing. Drift nonconvergence refers to whether the drift direction and drift rate of the peak trajectory converge within the steady-state allowable range, used to characterize whether frequency drift driven by slow variables is still ongoing. Spectral spike remnant refers to whether the spectral spike density is higher than the steady-state background and lacks a decay trend, used to characterize whether transient perturbations remain. To give the indicator a clear and computable form, the following four sub-results can be defined:

[0075]

[0076] in, The spectral window time synchronization flag is indicated as Node offset at the location; This indicates the current node distribution vector in the steady-state node category. The components can be taken as the proportion of node energy or the frequency of node occurrence; Indicates the steady-state spectral reference at the steady-state node category The reference component is used; the denominator is used to normalize the scale so that the identification of different installation locations is comparable; the formula describes the overall bias by summing the absolute values ​​of the differences between the category components of each node.

[0077]

[0078] in, The spectral window time synchronization flag is indicated as Abnormal edge residual degree at the location; This represents the set of edges in the current spectral pattern that are identified as newly added abnormal edges; This represents the set of all edges in the current spectral graph; Indicates that the endpoints in the current spectral pattern are and Edge weights; This represents the set of steady-state edges that are determined to be missing steady-state edges; Represents the set of steady-state edges in the steady-state spectral reference; This indicates the reference weight or reference intensity of the corresponding steady-state connection in the steady-state spectral reference. The value represents the penalty weight for missing edges, and is a non-negative number. The first term of this formula describes the weight ratio of newly added abnormal edges, and the second term describes the weight ratio of missing steady-state edges, so that the residual abnormal edges and the missing steady-state edges can be reflected at the same time.

[0079]

[0080] in, The spectral window time synchronization flag is indicated as The degree of non-convergence of the drift at that point; Indicated in the spectral window The set of available spectral peak trajectories nearby; Represents the spectral peak trajectory In the spectral window The drift rate at a given point can be obtained by dividing the difference in the center frequencies of the spectral peaks of adjacent spectral windows by the spectral window spacing. The drift rate reference value represents the steady-state spectral reference. This represents the drift tolerance and is a non-negative number. This formula describes whether the drift is still in a non-convergent state by statistically analyzing the proportion of drift rates that exceed the allowable drift range in steady state.

[0081]

[0082] in, The spectral window time synchronization flag is indicated as Spectral residue at the location; This indicates the spectral spike density of the current spectral window; The reference value for the spectral spike density represents the steady-state spectral pattern reference. The expression represents the spectral spike tolerance, and is taken as a non-negative number. This formula normalizes the spectral spike density beyond the steady-state background and tolerance, enabling the stable quantification of spectral spike residue. , , and By merging the results at the same spectral window time synchronization marker, the evaluation results are expressed on the time axis, thus providing a directly usable quantitative basis for subsequent determination of deviation persistence and marking of incomplete latent recovery based on a preset period.

[0083] S160. Under the premise that the external steady-state conditions are met, if it is determined that any sub-result in the evaluation results deviates from the steady-state spectral benchmark by a preset period, the corresponding recovery release segment is marked as a state of incomplete implicit recovery. In the subsequent steady-state re-entry segment, the secondary anomaly triggered by the load mutation is bound to the state of incomplete implicit recovery as a delayed recurrence homologous link based on the spectral inheritance relationship, so as to realize the identification and attribution of the implicit recovery failure caused by the incomplete release of the slow variable in the control loop.

[0084] Specifically, when reading the evaluation results and forming a sub-result sequence under the premise that the external steady-state conditions are met, the evaluation result entries whose time synchronization identifiers fall within the time synchronization identifier interval are first selected from the evaluation results based on the segment start time synchronization identifier and segment end time synchronization identifier of the recovery and release segment, and then spliced ​​into an evaluation result sequence in ascending order of time synchronization identifiers. Subsequently, the evaluation result sequence is grouped according to the installation position identifier, so that each installation position identifier corresponds to an evaluation result sub-sequence, and four sub-results, namely node bias, abnormal edge residual, drift non-convergence, and spectral spike residual, are extracted from the evaluation result sub-sequence, forming a node bias sub-result sequence, an abnormal edge residual sub-result sequence, a drift non-convergence sub-result sequence, and a spectral spike residual sub-result sequence that are bound one-to-one with the installation position identifier. External steady-state conditions refer to the externally observable steady-state premises obtained from the stability determination of DC bus voltage, DC bus current, and AC side phase current, which are used to ensure that the external conditions appear to have recovered. Evaluation results refer to the set of multi-dimensional quantitative results output by attractor regression determination. Sub-results refer to the single-dimensional quantitative components in the evaluation results. Sub-result sequence refers to the time series of sub-results arranged in order on the time synchronization identifier dimension. Installation location identifier is used to ensure that sub-results from different vibration pickup installation points are not mixed, so that subsequent deviation persistence determination and binding of the same source link can be completed in a closed loop on the same spatial object.

[0085] When performing deviation persistence determination and generating homology identifiers on sub-result sequences, a joint criterion of deviation intensity, deviation persistence, and hysteresis retention is first constructed for each sub-result sequence to avoid false alarms caused by short-term spikes, and deviation persistence is characterized by a preset period. Deviation intensity is used to measure the degree of deviation of the sub-result relative to the steady-state spectral benchmark, deviation persistence is used to measure the continuous existence of the deviation on the time synchronization identifier axis, and hysteresis retention is used to maintain the determination stability when the deviation is just approaching the threshold. When any sub-result sequence satisfies the deviation persistence determination, the corresponding recovery and release segment is marked as a hidden recovery incomplete state, and a homology identifier is generated, so that the homology identifier is bound to the installation location identifier and the time range of the recovery and release segment, ensuring the uniqueness and traceability of subsequent cross-segment associations. Steady-state spectral baseline refers to the reference structural acoustic baseline constructed and bound to the installation location identifier in the steady-state reentry segment; deviation persistence determination refers to the mechanism for determining whether the deviation is continuous for a preset period; the preset period refers to the minimum time length required for the deviation to be continuous, which can be expressed by the number of continuous time synchronization identifiers or the number of continuous spectral windows; latent recovery incomplete state refers to the state marker that the external steady-state conditions have been met but the internal structural acoustic state has not returned to the steady-state spectral baseline; homology identifier refers to the unique identifier used to bind a latent recovery incomplete state to a subsequent secondary anomaly as the same causal link. To make the deviation persistence determination robust, the sub-result sequence can be first converted into continuous confidence and the minimum cost persistence segment can be located:

[0086]

[0087] in, This indicates the time synchronization identifier index position of a certain sub-result sequence. The value at the location is given by any one of the node bias, abnormal edge residual, drift non-convergence, or spectral spike residual. This indicates the reference value or the center representative value of the reference range for this sub-result in the steady-state spectral reference. This indicates tolerance, used to allow for normal fluctuations near the steady-state spectral baseline; This represents the slope temperature parameter; a positive number and a smaller value indicate greater sensitivity near the threshold. The value represents the deviation confidence level, ranging from 0 to 1. This formula maps the degree of deviation from the reference value and tolerance to a continuous confidence level through a logical function, so that the deviation is not a binary abrupt change but a smooth growth, which facilitates subsequent continuous positioning.

[0088]

[0089] in, This represents the set of indices that were determined to be in the deviation category. This represents the set of indices indicating the optimal deviation. The set boundary complexity can be measured by the number of start and end boundaries of the deviation segment; The boundary penalty weight is a non-negative number, and the larger the value, the more it suppresses the fragmented deviation segments. This formula selects one or more deviation segments on the time axis in a minimum cost manner, so that the deviation confidence inside the deviation segment is as high as possible and the deviation confidence outside the segment is as low as possible. At the same time, it suppresses repeated switching caused by noise through boundary penalty.

[0090]

[0091] in, This represents the minimum continuous length corresponding to the preset period, and is a positive integer; when there exists a length not less than... The continuous interval is contained in When the sub-result is confirmed to meet the deviation persistence criterion, the generation of the implicit recovery incomplete state marker and the same-origin identifier is triggered.

[0092] After entering the steady-state reentry segment, when identifying load mutation events and backtracking to extract spectral pattern subsequences, the operating state sequence corresponding to the steady-state reentry segment is continuously read. Using the time synchronization identifier as an index, joint mutation detection is performed on the DC bus voltage, DC bus current, AC side phase current, and duty cycle sequences. After detecting a load mutation event, the time synchronization identifier corresponding to the trigger time of the load mutation event is recorded as the mutation time synchronization identifier. The preset observation window is backtracked with the mutation time synchronization identifier as the right endpoint to obtain the observation window time range. Subsequently, spectral patterns that fall within the observation window time range after the spectral window time synchronization identifier is projected onto the time synchronization identifier axis are selected from the spectral pattern sequence. They are then spliced ​​into spectral pattern subsequences in ascending order of the spectral window time synchronization identifiers, while maintaining a consistent association between the spectral pattern subsequences and the installation position identifiers. The steady-state reentry segment refers to the operational phase following the recovery exit event; the load mutation event refers to a rapid change in current and modulation caused by traction load changes, regenerative braking switching, or power dispatch switching during the period when external steady-state conditions are met; the preset observation window refers to the backtracking time range set to capture the potential residual state before the load mutation is triggered, and its length can be set according to the number of time synchronization markers; the spectral pattern subsequence refers to the sequence of spectral pattern image segments extracted from the steady-state reentry segment and used for homology determination. To enhance the stability of load mutation event identification, a multivariate normalized rate of change can be constructed and robust thresholding can be applied.

[0093]

[0094] in, Indicates the DC bus voltage. Indicates DC bus current. This represents the composite representative value of the AC phase current or the selected phase current. Represents the representative value of the duty cycle sequence; This represents the weight of each variable, can be a non-negative number, and is used to balance the contributions of different variables to mutation detection; It represents the level of fluctuation of the differenced series of variables on a robust scale, and is used to suppress noise scale differences; It is a stable term; This indicates the intensity of the combined mutation; a larger value indicates a higher likelihood of a load mutation event.

[0095] Furthermore, when the condition is met for the first time within Q consecutive time synchronization markers, k is determined to be the abrupt time synchronization marker. Wherein, Indicates the mutation threshold; The consistency length is used to avoid single-point false triggering. The principle of this criterion is to locate the moment of sudden change by using the normalized rate of change after amplification of multivariate consistency, and to use it as the anchor point of the backtracking observation window.

[0096] When determining and binding delayed recurrence homologous links based on spectral inheritance relationships, firstly, under the constraint of homologous identifiers already generated in the state of incomplete latent recovery, the spectral map at the end of the recovery release fragment corresponding to the state of incomplete latent recovery is located and used as the inherited source spectral map. Then, the consistency between the spectral map subsequences obtained by backtracking the load mutation event and the inherited source spectral map in terms of node category distribution, abnormal connection pattern, spectral peak drift shape, and spectral spike density level is calculated for each subsequence, and a comprehensive spectral inheritance consistency score is formed. When the spectral inheritance consistency score reaches the inheritance determination threshold and remains valid within a preset continuous window, the spectral inheritance relationship determination is confirmed, thereby binding the secondary anomaly triggered by the load mutation and the state of incomplete latent recovery as a delayed recurrence homologous link through homologous identifiers. Spectral inheritance relationship refers to the residual pattern of the latent incomplete recovery state in the key features of the structural acoustic state before the occurrence of subsequent secondary anomalies; secondary anomalies refer to abnormal manifestations such as voltage fluctuations, current-limited reentry, or control saturation recurrence triggered after a load mutation event; delayed recurrence homologous links refer to cross-segment causal link records connected by homologous identifiers, used to group the latent incomplete recovery state in the recovery release segment and the secondary anomaly in the steady-state reentry segment into the same root cause event. The spectral inheritance consistency score can be defined as a weighted combination of multi-component similarity, with higher weights given to anomaly connection patterns and drift morphologies to highlight the residue of slow variables.

[0097]

[0098] in, The spectral window time synchronization identifier in the spectral pattern subsequence is indicated by The consistency score of the spectral inheritance of the spectral pattern relative to the inherited source spectral pattern; This represents the node class distribution vector of the current spectral pattern in the steady-state node class space. Represents the node category distribution vector of the inherited source pattern map; This represents the abnormal edge pattern vector of the current spectral graph. The vector components can be encoded in the steady-state edge category space by the weight ratio of newly added abnormal edges and missing steady-state edges. Represents the vector of abnormal edge patterns inherited from the source spectrogram; This represents the drift rate value corresponding to the current spectral pattern. This represents the drift rate value corresponding to the inherited source spectrum pattern. This indicates the spectral spike density corresponding to the current spectral pattern. This indicates the density of spectral spikes corresponding to the inherited source spectral pattern. , , , This represents the weighting coefficient, which is non-negative and satisfies weight normalization to explain the contribution ratio; For the stable term; this formula maps the smaller the distribution difference, the smaller the pattern difference, the smaller the drift difference, and the smaller the spectral spike difference to similarity components around 0 to 1 and sums them by weight, so that the inheritance consistency score can comprehensively reflect the structural residue and dynamic residue.

[0099] Furthermore, the determination of spectral pattern inheritance is established only when there are R consecutive spectral window time synchronization flags. Indicates the threshold for inheritance determination; The inheritance duration is indicated to avoid misbinding caused by single-point coincidence similarity. When the spectral inheritance relationship is determined to be valid, the secondary anomaly is recorded as a delayed recurrence homologous link node corresponding to the homologous identifier. The time synchronization identifier range of the secondary anomaly occurrence, the corresponding installation location identifier, and the mutation time synchronization identifier that triggered the load mutation event are all written into the homologous link record, thereby completing the identification and attribution loop closure of the implicit recovery failure caused by the failure of the slow variable in the control loop to be released.

[0100] This application also provides a status monitoring device for subway flexible DC power supply equipment of an energy router, referring to... Figure 2 , Figure 2This is a schematic diagram of a status monitoring device for a subway flexible DC power supply equipment of an energy router, provided in an embodiment of this application. The device is a server, comprising an acquisition module 21 and a processing module 22. The DC-side filter inductor, AC-side parallel reactor, busbar film capacitor, and power module heat dissipation base of the energy router are pre-configured with vibration pickup units, and each vibration pickup unit is bound to a unique installation location identifier and a time synchronization identifier. The device includes an acquisition module and a processing module. The acquisition module 21 is used to acquire the vibration pickup sequence sent by the vibration pickup units and the operating status sequence bound to the energy router's operation during the operation of the energy router, and to correlate the vibration pickup sequence with the operating status sequence based on the time synchronization identifier. The state sequence is time-aligned to form an aligned sequence. The running state sequence includes a fault-crossing flag, a current-limiting flag, and a recovery flag. Processing module 22 is used to segment the aligned sequence based on the fault-crossing flag, current-limiting flag, and recovery flag to obtain target time segments. The target time segments include a drop-in segment, a current-limiting maintenance segment, a recovery-release segment, and a steady-state re-entry segment. Processing module 22 is also used to perform structural acoustic spectral generation processing on the target time segments to obtain spectral fragments, and to construct a spectral graph based on the spectral fragments to form a spectral graph sequence. The nodes of the spectral graph... The peak clusters in the spectral peak set represent the edges of the spectral pattern, and the energy transfer relationships reflected by the peak trajectories are used to represent the connections between peaks. Processing module 22 is also used to construct a steady-state spectral benchmark for each installation location marker in the steady-state reentry segment, based on the target spectral pattern sequence that has been determined to meet external steady-state conditions. The steady-state spectral benchmark is used to characterize the node distribution characteristics, edge density characteristics, peak drift characteristics, and spike density characteristics of the corresponding installation location marker under normal operating conditions. Processing module 22 is also used to perform attractor regression judgment on the spectral pattern sequence and the steady-state spectral benchmark in the recovery and release segment to obtain an evaluation. As a result, the evaluation results include abnormal edge residuals, node bias, drift non-convergence, and spectral spike residuals. The processing module 22 is also used to mark the corresponding recovery release segment as a hidden recovery incomplete state if any sub-result in the evaluation results deviates from the steady-state spectral benchmark by a preset period, provided that the external steady-state conditions are met. In subsequent steady-state re-entry segments, the secondary anomaly triggered by load mutation and the hidden recovery incomplete state are bound as delayed recurrence homologous links based on the spectral inheritance relationship, so as to realize the identification and attribution of the hidden recovery failure caused by the incomplete release of the slow variable in the control loop.

[0101] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0102] The communication bus 32 is used to enable communication between these components.

[0103] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0104] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0105] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0106] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a subway flexible DC power supply equipment status monitoring method using an energy router.

[0107] exist Figure 3 In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call the application program stored in the memory 35, which is a method for monitoring the status of a subway flexible DC power supply equipment of an energy router. When executed by one or more processors, the electronic device performs one or more methods as described in the above embodiments.

[0108] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0109] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for monitoring the status of a subway flexible DC power supply device using an energy router, characterized in that, The power router's DC-side filter inductor, AC-side parallel reactor, bus thin-film capacitor, and power module heat dissipation base are each pre-configured with vibration pickup units, and each vibration pickup unit is bound with a unique installation location identifier and time synchronization identifier. The method includes: During the operation of the energy router, the vibration pickup sequence sent by the vibration pickup unit and the operation status sequence bound to the operation process of the energy router are acquired, and the vibration pickup sequence and the operation status sequence are time-aligned based on the time synchronization identifier to form an aligned sequence. The operation status sequence includes a fault crossing flag, a current limiting flag, and a recovery flag. Based on the fault crossing flag, the current limiting flag, and the recovery flag, the alignment sequence is segmented to obtain a target time segment, which includes a drop-in segment, a current limiting maintenance segment, a recovery release segment, and a steady-state re-entry segment. Structural acoustic pattern generation processing is performed on the target time segment to obtain pattern fragments, and a pattern map is constructed based on the pattern fragments to form a pattern map sequence. The nodes of the pattern map are represented by peak clusters in the peak set, and the edges of the pattern map are represented by the energy transfer relationships reflected by the peak trajectories. In the steady-state reentry segment, based on the target spectral pattern sequence that has been determined to meet the external steady-state conditions, a steady-state spectral reference is constructed for each installation location identifier. The steady-state spectral reference is used to characterize the node distribution characteristics, edge density characteristics, spectral peak drift characteristics, and spectral spike density characteristics of the corresponding installation location identifier under normal operating conditions. Within the recovery and release segment, attractor regression is performed on the spectral pattern sequence and the steady-state spectral pattern benchmark to obtain evaluation results, which include abnormal edge residuals, node bias, drift non-convergence, and spectral spike residuals. Under the premise that the external steady-state conditions are met, if it is determined that any sub-result in the evaluation results deviates from the steady-state spectral benchmark by a preset period, the corresponding recovery release segment is marked as a state of incomplete implicit recovery. In the subsequent steady-state re-entry segment, the secondary anomaly triggered by the load mutation is bound to the state of incomplete implicit recovery as a delayed recurrence homologous link based on the spectral inheritance relationship, so as to realize the identification and attribution of the implicit recovery failure caused by the incomplete release of the slow variable in the control loop.

2. The method for monitoring the status of subway flexible DC power supply equipment using an energy router according to claim 1, characterized in that, During the operation of the energy router, the process of acquiring the vibration pickup sequence sent by the vibration pickup unit and the operating status sequence bound to the energy router's operation, and aligning the vibration pickup sequence and the operating status sequence based on the time synchronization identifier to form an aligned sequence, specifically includes: During the operation of the energy router, the time synchronization identifier and the installation location identifier are fixed for each vibration pickup unit, so that the time synchronization identifier uses the same timing reference and the same synchronization event source in the vibration pickup sequence and the operating state sequence. When acquiring the operating state sequence, the DC bus voltage, DC bus current, AC phase current, duty cycle sequence, fault ride-through flag, current limiting flag and recovery flag are synchronously sampled based on the synchronous event source, and a time synchronization identifier consistent with the vibration pickup sample is written for each operating state sampling point. On the data aggregation side, an alignment buffer is established based on the time synchronization identifier. Dual-stream sorting and association are performed on the vibration sequence and the running state sequence. When vibration samples and running state sampling points corresponding to the same time synchronization identifier are detected, alignment entries are generated to form the alignment sequence.

3. The method for monitoring the status of subway flexible DC power supply equipment using an energy router according to claim 1, characterized in that, The step of segmenting the aligned sequence into target time segments based on the fault-crossing flag, the current-limiting flag, and the recovery flag specifically includes: When segmenting the alignment sequence based on the fault crossing flag, the current limiting flag, and the recovery flag, the time synchronization flag corresponding to each alignment entry in the alignment sequence is read, and a flag sequence corresponding one-to-one with the alignment sequence is formed on the dimension of the time synchronization flag. The flag sequence includes the fault crossing flag, the current limiting flag, and the recovery flag. Perform flag debouncing processing on the flag sequence, and identify fault crossing entry event, rate limiting entry event, recovery entry event and recovery exit event after the flag debouncing processing; Based on the positions of the fault crossing entry event, the current limiting entry event, the recovery entry event, and the recovery exit event on the time synchronization identifier, the alignment sequence is divided into continuous intervals to obtain sequentially associated drop entry segment, current limiting maintenance segment, recovery release segment, and steady-state re-entry segment; After the segment division is completed, a segment consistency check is performed on each target time segment to verify whether the state continuity relationship of the fault crossing flag, the current limiting flag and the recovery flag in the corresponding segment meets the preset operational semantic constraints. When the segment consistency check fails, boundary correction processing is performed on the corresponding segment to reposition the segment start time synchronization flag and the segment end time synchronization flag to ensure that the target time segment is consistent with the alignment sequence in terms of time semantics.

4. The method for monitoring the status of subway flexible DC power supply equipment using an energy router according to claim 1, characterized in that, The process of performing structural acoustic pattern generation on the target time segment to obtain pattern fragments, and constructing pattern maps and forming pattern map sequences based on the pattern fragments, specifically includes: The vibration pickup sequence within the target time segment is subjected to vibration pickup preprocessing to obtain a preprocessed vibration pickup sequence. The vibration pickup preprocessing includes DC bias removal, amplitude normalization, bandwidth limitation, and pulse artifact suppression. Based on the preprocessed pickup sequence, short time window segmentation is performed to form spectral window segments, and a spectral window time synchronization identifier is bound to each spectral window segment; For each spectral window segment, peak set extraction is performed to obtain the peak set and peak energy distribution. Peak trajectory tracking is performed between adjacent spectral window segments to form peak trajectories. At the same time, the drift direction, drift rate and spike density of the peak trajectory are obtained. Based on the spectral peak set, a spectral peak cluster is constructed for each spectral window segment to obtain a spectral peak cluster, and the spectral window time synchronization identifier, the spectral peak cluster, the spectral peak trajectory, the drift direction, the drift rate and the spectral spike density are jointly encoded into a spectral pattern segment; A spectral pattern is constructed based on the spectral pattern fragments, and multiple spectral pattern images are spliced ​​together to form a spectral pattern sequence according to the time order of the spectral window time synchronization identifier, while maintaining a consistent association between the spectral pattern sequence and the target time segment type identifier.

5. The method for monitoring the status of subway flexible DC power supply equipment using an energy router according to claim 1, characterized in that, In the steady-state reentry segment, based on the target spectral pattern sequence that has been determined to meet the external steady-state conditions, a steady-state spectral reference is constructed for each installation location identifier, specifically including: In the steady-state reentry segment, the running state sequence corresponding one-to-one with the time synchronization identifier is read from the alignment sequence, and the fault crossing flag, current limiting flag and recovery flag are confirmed to be in an untriggered state in the running state sequence in order to determine the candidate interval that satisfies the flag consistency constraint. Within the candidate interval, the external steady-state conditions are determined based on the stability of the DC bus voltage, DC bus current, and AC side phase current, and the set of steady-state intervals is determined based on the external steady-state conditions. Based on the set of steady-state intervals, the spectral pattern sequence within the steady-state reentry segment is pruned to obtain a target spectral pattern sequence that is time-aligned with the set of steady-state intervals. For each installation location identifier, the corresponding set of spectral patterns is extracted from the target spectral pattern sequence, and node alignment and edge alignment processing are performed on the set of spectral patterns to form a set of stable node categories and a set of stable edge categories. Based on the set of steady-state node categories, node distribution characteristics are extracted; based on the set of steady-state edge categories, edge density characteristics are extracted; and spectral peak drift characteristics and spectral spike density characteristics are extracted simultaneously. After feature extraction is completed, steady-state spectral pattern screening is performed on the target spectral pattern sequence to remove abnormal spectral patterns. Based on the screened spectral pattern set, node distribution characteristics, edge density characteristics, spectral peak drift characteristics, and spectral spike density characteristics are reconstructed to build a corresponding steady-state spectral pattern benchmark for each installation location identifier. The steady-state spectral pattern benchmark is then bound and stored with the installation location identifier for subsequent attractor regression determination.

6. The method for monitoring the status of subway flexible DC power supply equipment using an energy router according to claim 1, characterized in that, Within the recovery and release fragment, attractor regression is performed on the spectral pattern sequence and the steady-state spectral benchmark to obtain the evaluation result, specifically including: Based on the time synchronization identifier, the spectral pattern sequence corresponding to the recovery and release segment is extracted, and the spectral pattern sequence is grouped according to the installation location identifier to form a spectral pattern sub-sequence that corresponds one-to-one with the steady-state spectral pattern reference. For each installation location identifier, the corresponding spectral pattern subsequence is structurally aligned with the steady-state spectral pattern reference. The structural alignment process includes mapping the spectral peak clusters in the spectral pattern to steady-state node categories, deviating steady-state node categories, or newly added abnormal node categories at the node level, and mapping the energy transfer relationships in the spectral pattern to steady-state connection categories, missing steady-state connections, or newly added abnormal connections at the connection level. After completing the structure alignment process, attractor regression analysis is performed on the spectral pattern subsequence along the spectral window time synchronization marker direction. The node distribution characteristics, edge density characteristics, spectral peak drift characteristics, and spectral spike density characteristics characterized by the steady-state spectral pattern benchmark are used as the target attractor state to track the regression process of the spectral pattern structure to the target attractor state. Based on the attractor regression analysis results, node bias is calculated at the node level, abnormal edge residual is calculated at the edge level, drift non-convergence is calculated at the spectral peak trajectory level, and spectral spike residual is calculated at the transient perturbation level, so that the node bias, abnormal edge residual, drift non-convergence, and spectral spike residual together constitute the evaluation result.

7. The method for monitoring the status of subway flexible DC power supply equipment using an energy router according to claim 1, characterized in that, Under the premise that the external steady-state conditions are met, if it is determined that any sub-result in the evaluation result deviates from the steady-state spectral benchmark by a preset period, the corresponding recovery release segment is marked as a state of incomplete implicit recovery. Furthermore, in subsequent steady-state re-entry segments, based on the spectral inheritance relationship, the secondary anomaly triggered by the load mutation is bound to the state of incomplete implicit recovery as a delayed recurrence homologous link. This achieves the identification and attribution of implicit recovery failure caused by the incomplete release of slow variables in the control loop, specifically including: Under the premise that the external steady-state conditions are met, the evaluation result sequence corresponding to the recovery release segment is read from the evaluation result based on the time synchronization identifier, and node bias degree sub-result sequence, abnormal connection residual degree sub-result sequence, drift non-convergence degree sub-result sequence and spectral spike residual degree sub-result sequence are formed respectively and bound one-to-one with the installation position identifier. The sub-result sequence is subjected to a deviation persistence determination. When any sub-result satisfies the deviation persistence determination, the corresponding recovery release segment is marked as an implicit recovery incomplete state and a homologous identifier bound to the installation location identifier and the recovery release segment time range is generated. After entering the steady-state reentry segment, the running state sequence corresponding to the steady-state reentry segment is continuously monitored to identify load mutation events. When a load mutation event is detected, the preset observation window before the load mutation event is triggered is traced back to extract the spectral pattern subsequence. Based on the consistency between the spectral pattern subsequence and the spectral pattern at the end of the incomplete latent recovery state in terms of node category distribution, abnormal connection pattern, spectral peak drift morphology, and spectral spike density level, a spectral pattern inheritance relationship determination is performed. When the spectral pattern inheritance relationship determination is successful, the secondary anomaly triggered by the load mutation and the incomplete latent recovery state are bound to the same source link of delayed recurrence through the same source identifier, thereby realizing the identification and attribution of the latent recovery failure caused by the incomplete release of the slow variable in the control loop.

8. A status monitoring device for subway flexible DC power supply equipment of an energy router, characterized in that, The device is used to execute the subway flexible DC power supply equipment status monitoring method of the energy router as described in any one of claims 1 to 7. The DC-side filter inductor, AC-side parallel reactor, bus thin-film capacitor, and power module heat dissipation base of the energy router are pre-configured with vibration pickup units, and each vibration pickup unit is bound with a unique installation location identifier and time synchronization identifier. The device includes an acquisition module and a processing module, wherein… The acquisition module is used to acquire the vibration pickup sequence sent by the vibration pickup unit and the operation status sequence bound to the operation process of the energy router during the operation of the energy router, and to time-align the vibration pickup sequence and the operation status sequence based on the time synchronization identifier to form an aligned sequence. The operation status sequence includes a fault crossing flag, a current limiting flag and a recovery flag. The processing module is used to segment the alignment sequence based on the fault crossing flag, the current limiting flag, and the recovery flag to obtain a target time segment. The target time segment includes a drop-in segment, a current limiting maintenance segment, a recovery release segment, and a steady-state re-entry segment. The processing module is further configured to perform structural acoustic pattern generation processing on the target time segment to obtain pattern fragments, and construct pattern maps based on the pattern fragments to form a pattern map sequence, wherein the nodes of the pattern map are represented by peak clusters in the peak set, and the edges of the pattern map are represented by the energy transfer relationships reflected by the peak trajectories; The processing module is further configured to, in the steady-state reentry segment, construct a steady-state spectral reference for each installation location identifier based on the target spectral pattern sequence that has been determined to meet the external steady-state conditions. The steady-state spectral reference is used to characterize the node distribution characteristics, edge density characteristics, spectral peak drift characteristics, and spectral spike density characteristics of the corresponding installation location identifier under normal operating conditions. The processing module is further configured to perform attractor regression determination on the spectral pattern sequence and the steady-state spectral pattern benchmark within the recovery and release segment to obtain evaluation results, the evaluation results including abnormal edge residual, node bias, drift non-convergence and spectral spike residual; The processing module is further configured to, under the premise that the external steady-state conditions are met, if it is determined that any sub-result in the evaluation result deviates from the steady-state spectral benchmark by a preset period, mark the corresponding recovery release segment as a state of incomplete implicit recovery, and bind the secondary anomaly triggered by the load mutation to the state of incomplete implicit recovery as a delayed recurrence homologous link based on the spectral inheritance relationship in the subsequent steady-state re-entry segment, so as to realize the identification and attribution of the implicit recovery failure caused by the incomplete release of the slow variable in the control loop.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images