Power distribution control system for a railway signal power supply system

Abstract

The present application belongs to the technical field of railway signal system, and discloses a power dispersion control system of a railway signal power supply system; by acquiring electrical parameters of each power module, energy-impedance coupling characteristics are calculated based on dissipation analysis, and state trajectory curvature is analyzed by using phase space theory to realize early and accurate identification of soft faults. By adopting selective removal and flexible disengagement strategies, undisturbed exit of the fault module is realized through step-by-step voltage reduction. An intelligent power redistribution mechanism is introduced, and the load is reasonably distributed according to the module stability margin ratio to prevent chain faults. Information sharing and energy storage pre-charging between controller groups are realized by a cooperative warning module to ensure smooth transition of the system. The adaptive correction function based on historical data enables the system to continuously optimize the judgment threshold and improve the identification accuracy. The present application effectively distinguishes internal faults from external load fluctuations, and significantly improves the reliability of the railway signal power supply system.

Description

Technical Field

[0001] This invention relates to the field of railway signaling system technology, and more specifically, to a power distributed control system for a railway signaling power supply system. Background Technology

[0002] Railway signal loads are characterized by significant complexity and variability, including instantaneous high-current surges during switch machine startup, inductive load characteristics of track circuits, sudden current demands during signal switching, and precise power supply requirements for computer interlocking equipment. Under such complex load conditions, existing protection mechanisms face fundamental challenges when power modules in the power system experience "soft faults" (e.g., aging of power electronic components, increased equivalent series resistance of filter capacitors leading to abnormal ripple, intermittent impedance changes caused by loose internal connections). Traditional control algorithms cannot effectively distinguish between normal load fluctuations and internal module faults, resulting in a dilemma: if the algorithm is adjusted to high sensitivity, instantaneous current changes during the startup of large equipment such as switch machines will be misinterpreted as faults, triggering unnecessary protection shutdowns, causing power outages to critical equipment, and affecting train scheduling; if the algorithm sensitivity is reduced to tolerate load fluctuations, the response to real faults will be delayed, allowing abnormal modules to continuously drag down the system bus voltage quality, causing voltage fluctuations, frequency instability, and harmonic interference. This instability could eventually trigger a protective restart of the entire station's computer interlocking monitoring system, impacting not only railway operational efficiency but also potentially jeopardizing train safety. Furthermore, current technology lacks the capability for "soft disconnection" of faulty modules, often employing a "hard disconnection" approach. Sudden disconnection of faulty modules under high load conditions can cause severe electrical transients, further deteriorating system stability.

[0003] In view of this, the present invention proposes a power distributed control system for railway signal power supply system to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a power distributed control system for a railway signal power supply system, comprising: The multi-source signal acquisition module is used to acquire the output current transient sequence and bus voltage ripple sequence of each power module, perform envelope extraction and high-frequency disturbance separation processing on the output current transient sequence, and generate current base waveform and high-frequency ripple echo attenuation factor. The dissipation analysis module is used to calculate the microscopic energy dissipation integral based on the phase shift relationship between the current base waveform and the bus voltage ripple sequence, and to construct the energy-impedance coupling spatial feature set by combining the high-frequency ripple echo attenuation factor. The reconstruction identification module is used to map the energy-impedance coupling space feature set to a multi-dimensional phase space, calculate the local manifold curvature of the current operating point trajectory, extract the phase space divergence feature value based on the local manifold curvature, and generate soft fault confidence. The selective cut-off module is used to divide the response interval according to the confidence level of soft faults. When it is in the fuzzy observation interval, it injects a tentative micro-perturbation into the corresponding power module and collects the response elastic recovery coefficient. Based on the elastic recovery coefficient, it generates a selective cut-off command. The flexible disconnect module is used to respond to the selective disconnection command, extract the current load rate of the target module, reduce the voltage setpoint of the target module in a step size that is inversely related to the current load rate, and monitor the bus voltage drop recovery rate after each voltage reduction until the output current is lower than the cutoff threshold and then perform physical disconnection. The redistribution module is used to assess the phase space stability margin of the remaining healthy modules during the physical disconnection of the target module, and to distribute the missing power to each healthy module according to the phase space stability margin ratio. The collaborative early warning module is used to encapsulate the soft fault confidence and selective disconnection command into a collaborative message and broadcast it to the distributed controller group to control adjacent modules to perform pre-charging operation of energy storage elements; The correction module is used to extract the normal vector offset of the phase space boundary surface based on historical load start feature data and erroneous cut-off records, and to reverse correct the divergent feature value judgment threshold in the reconstruction identification module.

[0005] The technical effects and advantages of the power distributed control system for railway signal power supply system of the present invention are as follows: This invention improves the accuracy of identifying soft faults in power modules of railway signaling power systems, effectively solving the technical challenge of fault diagnosis under complex load environments. The system can accurately distinguish between internal faults and external load fluctuations, significantly reducing the false alarm rate and avoiding power outages caused by accidental disconnections, while ensuring timely response to genuine faults. The adopted gradual power withdrawal strategy achieves disturbance-free fault isolation, eliminating the electrical shock risks associated with traditional switching methods. A distributed collaborative mechanism ensures intelligent load redistribution among modules, maintaining stable system operation and preventing fault propagation. Adaptive optimization capabilities enable the system to continuously improve fault diagnosis accuracy, enhancing long-term reliability in changing environments and providing continuous and stable power assurance for railway signaling equipment. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of a power distributed control system for a railway signal power supply system according to the present invention. Detailed Implementation

[0007] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0008] This application provides a power distributed control system for a railway signal power supply system. The system's execution entities include, but are not limited to, railway signal power supply control devices, distributed power management systems, and intelligent power monitoring platforms, which can be considered general control nodes in this application. The power control system includes, but is not limited to, at least one of the following: distributed power module groups, collaborative control networks, and intelligent fault diagnostic devices.

[0009] Please see Figure 1 In this embodiment of the invention, a power distributed control system for a railway signal power supply system includes: A multi-source signal acquisition module is used to acquire the output current transient sequence and bus voltage ripple sequence of each power module. It performs envelope extraction and high-frequency disturbance separation on the output current transient sequence to generate the current base waveform and high-frequency ripple echo attenuation factor. This module captures key electrical parameters of each power module in real time through a high-speed sampling circuit, including the time-domain fluctuation characteristics of the output current and the ripple variation characteristics of the bus voltage. The output current transient sequence records the current changes at the microsecond level, and the bus voltage ripple sequence reflects the subtle fluctuations in the bus voltage. Advanced signal processing algorithms are used to extract the envelope of the original sequences, separate high-frequency disturbance components, and obtain characteristic parameters that reflect the health status of the power modules. This data provides comprehensive raw material for subsequent analysis, ensuring the completeness and accuracy of fault diagnosis.

[0010] The dissipation analysis module calculates the microscopic energy dissipation integral based on the phase shift relationship between the current-based waveform and the bus voltage ripple sequence, and constructs an energy-impedance coupling spatial feature set by combining it with the high-frequency ripple return attenuation factor. The dissipation analysis module analyzes the operating status of the power module from the perspective of energy conversion efficiency, calculating energy loss characteristics through the phase relationship between the current and voltage waveforms. The microscopic energy dissipation integral describes the energy loss caused by non-ideal factors within the power module, while the high-frequency ripple return attenuation factor reflects the module's ability to suppress transient disturbances and its response characteristics. Together, they constitute the energy-impedance coupling spatial feature set, providing key indicators from an energy perspective for assessing the health status of the power module.

[0011] The reconstruction and identification module maps the energy-impedance coupling spatial feature set to a multi-dimensional phase space, calculates the local manifold curvature of the current operating point trajectory, extracts phase space divergence feature values ​​based on the local manifold curvature, and generates a soft fault confidence score. This module analyzes the dynamic behavior characteristics of the power module using phase space theory, transforming static features into a dynamic trajectory description. The mapped multi-dimensional phase space intuitively reflects the evolution trend of the system state, the local manifold curvature calculation reveals the nonlinear variation characteristics of the trajectory, and the phase space divergence feature values ​​quantify the degree to which the system state deviates from the stable region. The soft fault confidence score, as the final output, provides a quantitative basis for early fault identification, enabling the system to take preventative measures before the fault develops to a severe stage.

[0012] The selective clearance module is used to divide the response interval based on the confidence level of the soft fault. When the fault falls within the fuzzy observation interval, a tentative micro-perturbation is injected into the corresponding power module, and the response elastic recovery coefficient is collected. A selective clearance command is generated based on the elastic recovery coefficient. This module adopts a differentiated processing strategy according to different intervals of the soft fault confidence level. For cases where the confidence level is in the fuzzy interval, further diagnosis is performed by actively injecting micro-perturbations. The tentative micro-perturbation is a harmless small-signal excitation that can stimulate potential abnormal response characteristics of the system without affecting normal operation. The elastic recovery coefficient reflects the power module's response capability and recovery characteristics to perturbations and is a key indicator for distinguishing between internal soft faults and external load fluctuations. The selective clearance command generated based on accurate judgment avoids the risk of erroneous operation and improves system reliability.

[0013] The flexible disconnect module, in response to selective disconnection commands, extracts the current load rate of the target module and reduces the voltage setpoint of the target module in a stepped manner inversely correlated with the current load rate. After each voltage reduction, it monitors the bus voltage sag recovery rate until the output current falls below the cutoff threshold, at which point physical disconnection is performed. This flexible disconnect module implements a smooth power module exit mechanism, avoiding the electrical shocks associated with traditional direct disconnection methods. The stepped voltage reduction strategy dynamically adjusts the voltage reduction step size based on the module's current load; the heavier the load, the smaller the step size, ensuring the smoothness of the voltage reduction process. Monitoring the bus voltage sag recovery rate serves as a safety mechanism to prevent system instability during voltage reduction. The entire process achieves a "soft landing" for the power module, minimizing disturbances to system operation and improving the overall reliability of the power system.

[0014] The power redistribution module assesses the phase space stability margin of the remaining healthy modules during the physical disconnection of the target module and distributes the missing power to each healthy module proportionally based on the phase space stability margin. This module is responsible for redistributing power load after the faulty module is disconnected, ensuring continuous and stable system operation. Phase space stability margin is a key indicator for evaluating module load-bearing capacity, reflecting the safe distance of the module from the instability boundary. The proportional allocation strategy based on stability margin ensures the rationality of power redistribution and avoids the risk of cascading failures caused by overload of certain modules. This intelligent allocation mechanism maximizes the utilization of remaining system resources, maintains continuous power supply to critical loads, and improves system resilience and reliability.

[0015] The collaborative early warning module encapsulates soft fault confidence levels and selective clearing commands into collaborative messages and broadcasts them to the distributed controller group, controlling adjacent modules to perform pre-charging operations on energy storage elements. This module enables information sharing and coordinated actions among distributed controllers, transforming local fault information into system-level response strategies. The collaborative messages contain fault information and clearing plans, allowing other controllers to be aware of impending system changes in advance. The pre-charging operation of energy storage elements is a proactive mechanism that prepares for increased loads by adjusting the energy storage state in advance, reducing electrical transients during power redistribution, and improving the system's smooth transition capability and response speed.

[0016] The correction module, based on historical load start-up characteristic data and erroneous tripping records, extracts the normal vector offset of the phase space boundary surface and uses this offset to correct the divergence characteristic value judgment threshold in the reconstruction identification module. Through historical data analysis, the correction module continuously optimizes system parameters, achieving the system's self-learning and adaptive capabilities. Historical load start-up characteristics and erroneous tripping records provide valuable empirical data for the system, and the normal vector offset of the phase space boundary surface reflects the difference between the theoretical model and actual conditions. By using this data to correct the judgment threshold, the system can continuously improve the accuracy of fault identification, reduce the false positive rate, and achieve closed-loop optimization and continuous evolution of the control system.

[0017] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0018] In this embodiment of the invention, the detailed implementation steps for extracting the envelope and separating the high-frequency disturbance of the output current transient sequence to generate the current substrate waveform and the high-frequency ripple echo attenuation factor include: Extreme point detection is performed on the transient sequence of the output current. Local maxima are connected to generate the upper envelope, and local minima are connected to generate the lower envelope. Extreme point detection is a fundamental step in envelope extraction, constructing waveform boundaries by identifying peak and valley points in the current sequence. The detection process employs a sliding window comparison method. For each sampling point, its size relationship with points within a fixed range before and after it is compared. Points satisfying the local maximum or minimum condition are marked as extreme points. To improve detection accuracy, multi-scale analysis technology is used, performing detection simultaneously at different time windows, and determining the final extreme point location through consensus voting. The upper envelope connects all local maxima through cubic spline interpolation, while the lower envelope connects all local minima, forming the outer and inner boundaries of the current waveform, providing boundary constraints for subsequent base waveform extraction.

[0019] The mean sequence of the upper and lower envelopes is calculated as the current base waveform. The current base waveform reflects the main trend and low-frequency variation characteristics of the current, and is the core signal after eliminating high-frequency noise and transient fluctuations. The calculation method employs a simple and effective point-to-point averaging strategy, calculating the arithmetic mean of the values ​​of the upper and lower envelopes at each time point to form a smooth base waveform. The base waveform captures the main trend of current variation, filters out high-frequency disturbance components, and provides a reliable signal basis for subsequent energy dissipation analysis. The formula for calculating the base waveform is: ; in, For time points The current base waveform value, For the upper envelope at time point The value, The lower envelope at time point The value of .

[0020] High-frequency components exceeding a preset frequency threshold are extracted from the transient output current sequence. The amplitude attenuation ratio of these high-frequency components in adjacent zero-crossing intervals is calculated, and the average of these attenuation ratios is used as the high-frequency ripple return attenuation factor. High-frequency component extraction employs digital filtering technology, using a high-pass filter to separate signal components exceeding the preset frequency threshold (typically 2-3 times the system switching frequency). The filtered high-frequency components reflect high-frequency characteristics such as switching noise, harmonic interference, and transient oscillations in the system. Zero-crossing detection is used to locate the oscillation period of the high-frequency components, and linear interpolation improves the accuracy of zero-crossing location. The amplitude attenuation ratio calculation selects the peak amplitude ratio within adjacent zero-crossing intervals, reflecting the attenuation rate of the high-frequency components and serving as an important indicator of the system's damping characteristics. The average attenuation ratio across multiple intervals is used as the high-frequency ripple return attenuation factor, quantifying the system's ability to suppress high-frequency disturbances. A larger attenuation factor indicates better system damping and stronger disturbance rejection capability.

[0021] In this embodiment of the invention, the detailed implementation steps for constructing the energy-impedance coupling spatial feature set include: The first phase angle at the moment of current abrupt change in the current-based waveform and the second phase angle at the moment of voltage drop to the valley value in the bus voltage ripple sequence are extracted. Phase angle extraction is a key step in analyzing energy conversion efficiency, revealing the temporal relationship between current and voltage through precise location of key moments. The moment of current abrupt change is identified by calculating the first derivative of the current-based waveform, pinpointing the time point where the slope exceeds a threshold; the moment of voltage valley value is determined through local minima search. To improve the accuracy of time series analysis, wavelet transform is used for multi-scale edge detection to accurately capture the abrupt changes in electrical parameters. The first and second phase angles are represented in degrees, with the starting point of the standard reference period as zero, intuitively reflecting the relative relationship between current and voltage changes and providing a time-domain perspective for energy dissipation analysis.

[0022] The absolute value of the phase difference between the first and second phase angles is calculated, and then integrated over time within a preset integration window to obtain the microscopic energy dissipation integral. Phase difference analysis is a core method for evaluating energy conversion efficiency, reflecting energy loss by quantifying the degree of asynchrony between current and voltage changes. The absolute value of the phase difference is calculated using the circumferential distance formula to ensure continuity in the calculation near the boundary between 0° and 360° of the phase angle. The time integration uses the trapezoidal rule, accumulating the phase difference contribution within a preset integration window (usually 3-5 times the system characteristic time constant) to form the microscopic energy dissipation integral. This integral value directly reflects the energy loss caused by phase mismatch in the system and is an important indicator of the power module's health status. An abnormally large increase in the integral value usually indicates aging of internal components or parameter drift. The formula for calculating the energy dissipation integral is as follows: ; in, For the integral of microscopic energy dissipation, For time points The first phase angle, For time points The second phase angle, The starting time of the integration. Set the preset width for the integration window.

[0023] An energy-impedance coupling space feature set is constructed by orthogonally combining the microscopic energy dissipation integral and the high-frequency ripple echo attenuation factor. Orthogonal combination is an effective method for constructing a multidimensional feature space, improving expressive power by ensuring low correlation between features. The combination process first normalizes the microscopic energy dissipation integral and the high-frequency ripple echo attenuation factor to eliminate dimensional differences; then, principal component analysis is used to ensure the orthogonality of feature directions and reduce information redundancy; finally, an energy-impedance coupling space feature set is formed as a comprehensive characterization of the module's health status. This feature set describes the power module's operating state from two key dimensions: energy conversion efficiency and dynamic impedance characteristics, providing a high-quality feature foundation for subsequent multidimensional phase space mapping and trajectory analysis.

[0024] In this embodiment of the invention, the detailed implementation steps for generating soft fault confidence include: A three-dimensional phase space is constructed using the dissipation integral of the energy-impedance coupling spatial feature set as the first coordinate axis, the return attenuation factor as the second coordinate axis, and the time derivative as the third coordinate axis. The construction of the three-dimensional phase space is a fundamental step in state trajectory analysis, characterizing the core dynamic characteristics of the system by selecting appropriate coordinate axes. The coordinate axis design follows the principles of orthogonality and physical meaning: the first coordinate axis reflects the energy loss state, the second axis characterizes the impedance dynamic characteristics, and the third axis captures the rate of state change. The central difference method is used to calculate the time derivative, improving the accuracy of derivative estimation. The constructed three-dimensional phase space provides a unique spatial location mapping for the system state at each moment. The temporal evolution of the system state manifests as a continuous trajectory in space, intuitively reflecting the dynamic behavior characteristics of the power module and providing a geometric perspective for the early identification of abnormal states.

[0025] In three-dimensional phase space, the curvature of the principal normal vector of the trajectory curve formed by the current operating point and a preset number of previous operating points is calculated and denoted as the local manifold curvature. Local manifold curvature is an important measure of the trajectory's nonlinear characteristics, reflecting the regularity of system state changes by calculating the degree of trajectory curvature. The calculation process first selects the current operating point and a preset number (usually 5-10) of previous operating points to form a local trajectory segment; then, differential geometry methods are used to estimate the principal normal vector and curvature of the trajectory. To improve computational stability, the moving least squares method is used to fit the local trajectory, and then differential geometric quantities are calculated based on the fitted curve. The local manifold curvature directly reflects the acceleration characteristics of system state changes; an abnormally large increase in curvature usually indicates a sudden change in system behavior, possibly foreshadowing early signs of a fault. The curvature calculation formula is: ; in, For time points The local manifold curvature, and The trajectories at different time points are respectively The first and second derivative vectors, Represents the cross product of vectors. This represents the Euclidean norm of a vector.

[0026] The rate of change of variance of the local manifold curvature within a sliding observation window is calculated as a phase space divergence eigenvalue. The phase space divergence eigenvalue is a quantitative indicator of system stability, reflecting the increase in uncertainty in system behavior by assessing changes in the statistical properties of curvature. The calculation process first sets an appropriate sliding observation window (typically 2-3 times the system response time), calculates the variance of the local manifold curvature sequence within the window, and then compares the variance values ​​of adjacent windows to calculate the rate of change of variance. The rate of change of variance can effectively capture abrupt changes in the dynamic behavior of the system, exhibiting high sensitivity to early, small changes, making it an ideal indicator for fault early warning. The calculation of the rate of change of variance employs exponential smoothing to reduce the impact of random fluctuations and improve the reliability of trend identification.

[0027] When the phase space divergence eigenvalue exceeds the baseline divergence interval, a soft fault confidence score is generated based on the excess proportion. The soft fault confidence score is a quantitative assessment of the system's fault probability, providing an intuitive risk representation by converting technical indicators into confidence levels. The generation process first determines the baseline divergence interval, typically set based on historical data statistics and expert knowledge to define the typical divergence range of normal operating conditions. Then, it calculates the proportion by which the current divergence eigenvalue exceeds the upper limit of the interval, converting this excess proportion into a confidence score value within the [0,1] interval using a nonlinear mapping function. The mapping function employs an S-curve design, exhibiting high sensitivity in the critical region and effectively distinguishing boundary states. As a key output for early fault identification, the soft fault confidence score directly impacts subsequent decision-making strategies, providing a quantitative basis for selective module removal. The confidence score calculation formula is as follows: ; in, For soft fault confidence, For the phase space diverging eigenvalues, and These are the lower and upper limits of the baseline divergence interval, respectively. For sigmoid mapping functions, the Sigmoid function is typically used: ,in This is the curve steepness parameter.

[0028] In this embodiment of the invention, the detailed implementation steps for generating selective resection instructions include: The confidence level of soft faults is compared with preset lower and higher thresholds. Values ​​below the lower threshold are considered healthy, values ​​above the higher threshold are considered confirmed faults, and values ​​in between are considered fuzzy observation intervals. This response interval division is the foundation of the tiered processing strategy, allowing for differentiated handling of different risk levels by setting appropriate thresholds. The lower threshold (typically set at 0.3-0.4) defines minor risks that can be safely ignored, while the higher threshold (typically set at 0.7-0.8) defines definitive faults requiring immediate attention. Threshold settings are based on extensive historical operational data analysis and system reliability requirements, balancing sensitivity and false alarm rate. The fuzzy observation interval represents a state of particular concern, indicating signs of a fault that have not yet been definitively confirmed, requiring further diagnosis to avoid misjudgment. This three-interval division strategy effectively avoids the limitations of traditional binary judgments and improves the system's decision-making intelligence.

[0029] When the system is in a fuzzy observation range, a voltage micropulse with a preset duty cycle width is superimposed on the pulse width modulator of the corresponding power module. Voltage micropulse injection is an active diagnostic technique that observes the system's response characteristics by introducing known, minute perturbations. The pulse design principle is to be small enough not to affect normal operation, yet large enough to trigger potential anomalies. The pulse width is typically set to 5%-10% of the modulation period, and the pulse amplitude is controlled within 1%-3% of the rated voltage to ensure that the diagnostic process does not affect system stability. The micropulse is injected through an additional control channel of the modulator, achieving precise synchronization with the normal control signal and avoiding timing conflicts. This micro-perturbation method is similar to stress testing in medical diagnosis, amplifying the symptoms of potential problems through slight stimulation, thereby improving the accuracy and reliability of the diagnosis.

[0030] The time constant required for the output current to return to the baseline after the voltage pulse disappears is collected, and the reciprocal of this time constant is defined as the elastic recovery coefficient. The elastic recovery coefficient is an important indicator of the system's dynamic response capability, reflecting its internal health by quantifying the speed at which the system recovers from a disturbance. The time constant measurement employs an exponential decay model fitting method, performing least-squares fitting on the current waveform after the pulse disappears to extract the characteristic time constant. Normally functioning power modules exhibit a high elastic recovery coefficient and can quickly recover to a stable state; while modules with soft faults typically show slow recovery, with a significantly reduced elastic recovery coefficient. The formula for calculating the elastic recovery coefficient is: ; in, The elastic recovery coefficient, The time constant required for the current to fall back to the baseline is obtained through exponential decay fitting: ,in Baseline current, The amplitude of the current change caused by the pulse. This is the moment the pulse disappears.

[0031] When the elastic recovery coefficient is lower than the preset elastic lower limit, it is determined to be an internal soft fault, and a selective clearance command is generated; otherwise, it is determined to be an external inductive load inrush, and the clearance operation is blocked. Distinguishing between soft faults and external interference is a key step in avoiding erroneous actions, and accurate judgment is achieved through differences in elastic recovery characteristics. The preset elastic lower limit is determined based on the statistical distribution of normal modules, typically set as the lower quartile of the elastic recovery coefficient distribution under normal operating conditions. The judgment logic uses a threshold comparison method. When the measured value is below the threshold, the system has sufficient confidence to consider that the module has an internal soft fault and needs to perform a clearance operation; when the measured value is above the threshold, the system considers the observed anomaly to be a temporary phenomenon caused by external load characteristics and does not require clearance. This differentiation method based on dynamic response characteristics greatly improves the accuracy of fault identification, effectively avoids the erroneous clearance of normally operating modules, and improves the stability and availability of the system.

[0032] In this embodiment of the invention, in response to a selective cut-off command, the current load rate of the target module is extracted, the voltage setpoint of the target module is reduced in a step size inversely related to the current load rate, and the bus voltage drop recovery rate is monitored after each voltage reduction until the output current is lower than the cutoff threshold before physical disconnection is performed. The detailed implementation steps include:

[0033] The selective disconnection command is parsed to obtain the target module identifier. The ratio of the current output power to the rated power corresponding to the target module identifier is read as the current load factor. Load factor calculation is the foundation of the flexible disconnection strategy, determining the appropriate disconnection speed by accurately assessing the module's operating status. Command parsing adopts a standard message processing flow to extract the unique identifier and additional control parameters of the target module. The current output power is calculated by multiplying the voltage and current, while the rated power is read from the module configuration database. The ratio of the two directly reflects the module's load level. The load factor calculation considers real-time efficiency correction, dynamically adjusting the rated power based on temperature and operating time to ensure that the calculation results accurately reflect the module's actual state. This load assessment based on real-time data provides a scientific basis for subsequent tiered derating and is a key prerequisite for achieving smooth disconnection.

[0034] A stepped derating step size is set, which is inversely proportional to the current load rate. Stepped derating is the core strategy for achieving smooth disconnection, reducing the impact of the disconnection process by gradually decreasing the output voltage. The step size design follows the principle of inverse proportionality with the load rate, using a small step size under high load conditions and a large step size under low load conditions, ensuring that the disturbance to the system caused by each derating is controlled within an acceptable range. The inverse relationship is achieved through a power function, which ensures a smooth transition under high load conditions and avoids unnecessary delays under low load conditions. The step size calculation also considers the dynamic response characteristics of the system, adaptively adjusting according to the bus capacitance value and load characteristics to ensure system stability during the derating process. This intelligent step size design significantly improves the adaptability and reliability of flexible disconnection, and can cope with disconnection requirements under various load conditions.

[0035] The voltage setpoint of the target module is progressively reduced in stepwise derating increments, and the bus voltage sag recovery rate is calculated within a set time window after each reduction. This progressive voltage reduction is a flexible disconnection process, ensuring safety by controlling each step and monitoring the system response. The voltage setpoint adjustment is implemented through a digital control interface, ensuring precise control and a smooth transition. After each adjustment, the system enters a monitoring phase, collecting dynamic response data of the bus voltage and calculating voltage sag and recovery characteristics. The recovery rate calculation uses a linear regression method to estimate the voltage slope during the recovery phase, intuitively reflecting the system's dynamic stability. The monitoring time window is set to 3-5 times the system's characteristic time constant to ensure the complete capture of the dynamic response process. This closed-loop control strategy of gradual adjustment and monitoring ensures the system remains stable throughout the voltage reduction process, avoiding the electrical shocks associated with traditional hard switching methods.

[0036] When the drop recovery rate falls below the safe recovery baseline, the next step of voltage reduction is paused, and the current voltage setpoint is maintained until the recovery rate meets the requirements, at which point further reduction can be resumed. Recovery rate monitoring is a safety mechanism for the flexible disconnection process, dynamically adjusting the voltage reduction pace to prevent system instability. The safe recovery baseline is determined based on system design parameters and operational experience, typically set to 50%-70% of the recovery rate under normal operating conditions. When an abnormally low recovery rate is detected, the system automatically pauses the voltage reduction operation, maintaining the current state until the system regains normal responsiveness. During the pause, the system continuously monitors key parameters and assesses the recovery trend of system stability. Subsequent voltage reduction steps are only resumed after the recovery rate once the requirements are met again. This adaptive control strategy significantly improves the safety of the disconnection process, ensuring stable system operation even under extreme load conditions and avoiding the risk of system crash due to excessively rapid operation.

[0037] When the target module's output current drops to a preset cutoff threshold, the physical contactor is disconnected to complete the physical disconnection. Physical disconnection is the final step in the flexible disconnection process, performing physical isolation after electrical parameters have dropped to a safe level. The cutoff threshold is set to 5%-10% of the rated current to ensure that the current during disconnection is small enough to avoid significant arcing and electrical shock. Precise timing control of the physical contactor ensures that the disconnection operation is only performed after the current has stabilized below the threshold, avoiding malfunctions caused by jitter. During the disconnection process, the system continues to monitor the bus voltage and system stability, preparing for possible anomalies. Immediately after physical disconnection, the system initiates a fault recording program to save a complete event sequence and parameter change history, providing data support for subsequent analysis and improvement. This "soft-then-hard" graded disconnection strategy minimizes the impact of faulty module removal on the system, achieving a smooth, disturbance-free exit, representing a significant innovation in system reliability design.

[0038] In this embodiment of the invention, during the physical disconnection of the target module, the detailed implementation steps of assessing the phase space stability margin of the remaining healthy modules and allocating the missing power to each healthy module according to the phase space stability margin ratio include: The system acquires the multidimensional distance from the current operating point of each healthy module to the boundary of a preset stable attractor in three-dimensional phase space, defining this multidimensional distance as the phase space stability margin. Stability margin calculation is the theoretical basis for power redistribution, assessing the module's load-bearing capacity through phase space analysis. The calculation process first determines the boundary surface of the stable attractor based on module parameters and operating conditions; this surface defines the range of stable module operation. Then, it measures the minimum distance from the current operating point to the boundary as a quantitative representation of the stability margin. The distance calculation uses an improved Mahalanobis distance, considering the physical meaning and weight differences of different dimensions, thus improving the accuracy of the margin assessment. To improve computational efficiency, the system employs an approximation algorithm and local linearization technique to achieve real-time margin assessment. The phase space stability margin intuitively reflects the module's available capacity and operational risk, providing a theoretical basis for scientific load allocation and ensuring that the redistribution process does not cause any module to exceed its safe operating range.

[0039] The phase space stability margin of each healthy module is normalized to obtain the phase space stability margin ratio. The stability margin ratio is the basis for fair allocation, and normalization enables capacity comparison between different modules. The normalization process uses a summation standardization method, dividing the stability margin of each module by the sum of the margins of all modules to obtain the ratio value in the interval [0,1], ensuring that the sum of the allocation coefficients is 1. To improve the rationality of the ratio allocation, the system introduces a nonlinear mapping mechanism, performing a power function transformation on the original margin values, increasing the weight of high-margin modules, and further optimizing the balance of load distribution. The normalization calculation also considers the historical stability and reliability records of the modules, fine-tuning the allocation ratio by introducing historical performance coefficients to ensure that the system prioritizes the use of more reliable modules, improving the overall system stability and reliability.

[0040] The system calculates the missing power carried by the physically disconnected module and allocates it to the corresponding healthy modules according to the phase space stability margin ratio, generating output current correction commands for each healthy module. Power allocation is the core step in the redistribution process, achieving seamless load transfer through precise calculation and control command generation. The missing power calculation is based on real-time monitoring data before disconnection, considering the power change trend during the voltage reduction process to ensure that the calculation results accurately reflect the actual gap. The allocation process strictly follows the stability margin ratio, with the incremental power allocated to each healthy module proportional to its relative margin. The allocation algorithm also includes a dynamic adjustment mechanism, fine-tuning the allocation coefficients based on real-time feedback during the allocation process to ensure system stability during redistribution. The output current correction command includes target current increment and transition time parameters, achieving a smooth transition through a ramp function to avoid electrical shocks caused by abrupt changes. This intelligent allocation strategy based on theoretical analysis maximizes the utilization of system resources while ensuring the stability and safety of the load transfer process.

[0041] In this embodiment of the invention, the detailed implementation steps for encapsulating the soft fault confidence and selective clearance command into a cooperative message and broadcasting it to the distributed controller group to control adjacent modules to perform energy storage element pre-charging operations include: The module identifier of the soft fault, the soft fault confidence level, and the derating progress status are encapsulated into a collaborative message. This collaborative message encapsulation is the foundation for information sharing among distributed controllers, enabling effective communication through a standardized data format. The encapsulation process employs a layered design: the core layer contains necessary identification and status information, while the extension layer contains optional detailed data and control parameters. The module identifier uses a system-unique identifier to ensure accurate location within complex networks; the soft fault confidence level is represented as a floating-point number, providing a quantitative assessment of the fault probability; the derating progress status includes the current step index and the total number of steps, reflecting the execution status of the decoupling process. The collaborative message also includes a timestamp and verification code to ensure data timeliness and integrity. A compact binary encoding format is used to minimize communication load and improve system response speed and communication efficiency.

[0042] The system broadcasts cooperative messages to adjacent controllers in the distributed controller group via an internal bus. This broadcasting is a crucial step in achieving system-level coordination, transmitting control information through an efficient and reliable communication network. The broadcast process employs optimized network topology and routing strategies to ensure messages reach all relevant controllers promptly. The internal bus utilizes redundancy and error detection mechanisms to enhance communication reliability and fault tolerance. To guarantee real-time performance, a priority scheduling strategy assigns the highest priority to fault-related messages, ensuring timely delivery even under high network load. The broadcast mechanism also includes message acknowledgment and retransmission mechanisms. Successful message delivery is verified by receiving acknowledgment signals, and automatic retransmission is performed for targets that have not received acknowledgment, ensuring the consistency and integrity of system information. This efficient and reliable communication mechanism provides a solid foundation for the cooperative actions of distributed controllers and is a key support for achieving system-level intelligent response.

[0043] Neighboring controllers parse coordination messages. When the soft fault confidence level exceeds the warning threshold, they calculate the pre-allocated power increment their own modules will bear. Coordination message parsing is the first step in response preparation, formulating subsequent action plans through a correct understanding of the received information. The parsing process first verifies the integrity and validity of the message to ensure data integrity; then it extracts key information, paying particular attention to the soft fault confidence level and derating status; finally, it determines whether a response is needed based on local configuration. The warning threshold is typically set to 0.6-0.7; preventative preparation is triggered when the confidence level exceeds this value. The pre-allocated power increment calculation is based on the system topology and load distribution strategy, considering the physical connections between modules and historical load characteristics, and estimating the additional load this module might bear after the faulty module is disconnected. The calculation results serve as a reference for pre-charging operations, ensuring that prepared resources match potential demand, improving the system's predictability and response efficiency.

[0044] Based on the pre-allocated power increment, the pre-excitation current of the internal filter inductor or the pre-charge voltage of the capacitor is adjusted in advance, enabling the energy storage elements to enter an energy pre-reservation state. Energy storage pre-charging is the core response of the coordinated early warning system, improving the system's ability to cope with sudden changes by preparing energy buffers in advance. For inductor energy storage, the system increases the pre-excitation current through fine-tuning the control loop to increase the stored energy in the inductor; for capacitor energy storage, the system appropriately increases the pre-charge voltage to increase the energy reserve of the capacitor. The pre-charge level is proportional to the pre-allocated power increment, ensuring that the reserved energy is sufficient to meet the instantaneous demand during load transfer. The operation process employs a gradual control strategy, using small-step adjustments to avoid disturbances to the system caused by the pre-charging process itself. The system simultaneously monitors the temperature and pressure parameters of the energy storage elements to ensure that the pre-charging operation does not exceed safe limits. This advance preparation mechanism significantly improves the system's ability to cope with sudden changes, reduces electrical transients during load transfer, and ensures the continuous and stable operation of the power system during fault handling.

[0045] In this embodiment of the invention, the detailed implementation steps for extracting the normal vector offset of the phase space boundary surface based on historical load start-up characteristic data and erroneous cut-off records, and for reversing and correcting the divergence characteristic value determination threshold in the reconstruction identification module, include: This process extracts records of erroneous events from historical data that were initially identified as faulty but ultimately confirmed as normal load startups. Erroneous event analysis is a crucial step in system self-optimization, improving judgment accuracy by learning from historical experience. The data extraction process first filters system logs for event records initially identified as faulty but ultimately verified as normal; these records contain complete parameter sequences and event tags. Then, the records are cleaned and standardized to eliminate noise and outliers. Finally, they are categorized and organized according to specific patterns and load types to form a structured erroneous event dataset. A multi-condition matching algorithm is used for data filtering to ensure the representativeness and analyzability of the extracted erroneous events. This learning method based on historical data provides real-world case studies to support system optimization and serves as a vital data foundation for adaptive evolution.

[0046] The distribution of the working point trajectories recorded in the three-dimensional phase space of malfunction events is extracted and fitted to generate the actual boundary surface. Boundary surface fitting is a key step in optimizing the judgment criteria, and statistical analysis is used to determine the actual boundaries between normal and abnormal conditions. The extraction process first maps the parameter sequences of malfunction events to three-dimensional phase space to reproduce the working point trajectories; then, density clustering is used to identify the distribution characteristics of the trajectories; finally, machine learning algorithms such as support vector machines or multilayer perceptrons are used to fit the distribution boundaries to generate the actual boundary surface. The fitting process employs cross-validation and regularization techniques to improve the generalization ability and stability of the surface. The actual boundary surface intuitively reflects the normal variation range of system parameters, providing a data-driven scientific basis for optimizing the judgment threshold and serving as an important reference benchmark for achieving accurate fault judgment.

[0047] The deviation of the angle between the normal vectors of the actual boundary surface and the preset theoretical boundary surface at each sampling point is calculated, and the mean of the deviation is used as the normal vector offset. Normal vector analysis is a precise method for quantifying model deviation, assessing the degree of deviation of the judgment criterion by measuring the directional differences of the boundary surfaces. The calculation process first uniformly samples the boundary surface to obtain a series of representative points; then, the normal vectors of the actual and theoretical surfaces are calculated at each sampling point; finally, the angle between the normal vectors at the corresponding point is calculated to quantify the directional difference between the two surfaces at that point. The selection of sampling points considers the complexity and curvature distribution of the boundary surface, increasing the sampling density in high curvature regions to ensure that the analysis results accurately reflect the characteristics of complex regions. The normal vector offset serves as the main basis for the correction of the judgment threshold, intuitively reflecting the systematic deviation between the theoretical model and the actual situation, and providing a scientific quantitative basis for subsequent correction operations.

[0048] The normal vector offset is multiplied by the damping attenuation coefficient, added to the current divergent characteristic value judgment threshold, and written into the reconstruction and identification module. Threshold correction is the final step in closed-loop optimization, improving the system's accuracy by reasonably adjusting the judgment criteria. The correction process first introduces the damping attenuation coefficient to control the magnitude and speed of correction, avoiding system instability caused by over-adjustment; then, the correction amount is added to the current threshold to generate a new judgment criterion that better reflects the actual situation; finally, the corrected threshold is written into the system configuration, completing the closed-loop optimization. The damping attenuation coefficient is typically set to 0.3-0.7, dynamically adjusted based on the system's historical stability and current operating status to ensure the smoothness and controllability of the correction process. Threshold correction adopts an incremental update strategy, gradually approaching the optimal value through multiple small adjustments, avoiding the risks that may arise from a single large adjustment. This adaptive optimization mechanism based on actual data enables the system to continuously learn and evolve, continuously improving the accuracy and reliability of fault diagnosis.

[0049] In this embodiment of the invention, after the system controls adjacent modules to perform pre-charging operations on energy storage elements, it further includes: The system monitors the overshoot amplitude of the bus voltage at the instant the target module physically disconnects. Overshoot monitoring is a crucial part of system performance evaluation, assessing operational effectiveness by recording changes in key parameters. The monitoring process employs high-speed data acquisition technology to accurately capture voltage fluctuations at the moment of physical disconnection, with a typical sampling rate of 100,000-500,000 points per second, ensuring accurate recording of microsecond-level transient processes. Overshoot amplitude calculation is based on a standardized definition, using the pre-disconnect steady-state voltage as a benchmark, and measuring the ratio of the peak voltage after disconnection to the benchmark. To improve measurement accuracy, the system uses digital filtering and peak detection algorithms to eliminate noise interference and accurately identify the true peak value. The monitoring data is statistically processed to generate an evaluation report, serving as important feedback information for system performance and providing empirical evidence for subsequent optimization. This refined performance monitoring mechanism provides a data foundation for continuous system improvement and is a vital support for ensuring long-term reliable operation.

[0050] When the voltage overshoot amplitude exceeds the preset overshoot tolerance range, the ratio of the voltage overshoot amplitude to the upper limit of the overshoot tolerance range is calculated. Overshoot analysis is a fundamental step in performance optimization, determining the direction and magnitude of improvement by quantitatively assessing the degree of overshoot. The tolerance range is set based on system design specifications and load requirements, typically defined as ±3%-5% of the rated voltage. Overshoot assessment employs standardized processing, calculating the relative proportion exceeding the tolerance upper limit, providing a unified performance metric. The analysis process also examines the duration and decay characteristics of the overshoot, comprehensively evaluating the quality of the transient process. Overshoot cause analysis uses pattern recognition technology, comparing historical cases and theoretical models to identify key factors leading to overshoot, providing guidance for targeted optimization. This closed-loop evaluation mechanism based on measured data enables the system to objectively evaluate its own performance, promptly identify areas for improvement, and provide a reliable basis for continuous optimization.

[0051] The system reduces the step size of the stepped flexible disengagement module during the next derating operation according to the overshoot ratio. Step size optimization is a concrete implementation of the system's self-adjustment, improving the smoothness of operation through parameter fine-tuning. The optimization strategy is based on feedback control principles, dynamically adjusting subsequent parameters according to the current operation effect to form a closed-loop control structure. The step size adjustment adopts a proportional relationship; the more severe the overshoot, the greater the step size reduction, ensuring the rationality and effectiveness of the adjustment. The system also introduces a learning factor to record the cumulative effect of multiple adjustments, avoiding repeated adjustments in the same direction and improving the convergence speed of optimization. The adjusted step size parameters are stored in non-volatile memory, ensuring that the system can inherit optimization experience after restarting and achieve long-term evolution. This adaptive control mechanism enables the system to continuously optimize its operation strategy according to actual operating conditions, constantly improving operational quality, and is an important manifestation of the system's intelligence and self-optimization.

[0052] This invention achieves early detection and seamless removal of soft faults in the power modules of railway signaling power supply systems through multi-source signal acquisition, dissipation analysis, reconstruction identification, selective removal, flexible disconnection, power redistribution, collaborative early warning, and threshold correction. The distributed control architecture of this invention can accurately identify potential faults and achieve smooth removal, effectively avoiding the impact of single-point faults on the system.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0054] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0055] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A power distributed control system for a railway signal power supply system, characterized in that, include: The multi-source signal acquisition module is used to acquire the output current transient sequence and bus voltage ripple sequence of each power module, and to perform envelope extraction and high-frequency disturbance separation processing on the output current transient sequence to generate current base waveform and high-frequency ripple echo attenuation factor. The dissipation analysis module is used to calculate the microscopic energy dissipation integral based on the phase shift relationship between the current base waveform and the bus voltage ripple sequence, and to construct an energy-impedance coupling spatial feature set in combination with the high-frequency ripple echo attenuation factor. The reconstruction identification module is used to map the energy-impedance coupling spatial feature set to a multi-dimensional phase space, calculate the local manifold curvature of the current operating point trajectory, extract phase space divergence feature values ​​based on the local manifold curvature, and generate soft fault confidence. The selective cut-off module is used to divide the response interval according to the soft fault confidence. When it is in the fuzzy observation interval, it injects a tentative micro-perturbation into the corresponding power module and collects the response elastic recovery coefficient. Based on the elastic recovery coefficient, it generates a selective cut-off command. The flexible disconnect module is used to respond to the selective disconnection command, extract the current load rate of the target module, reduce the voltage setpoint of the target module in a step size that is inversely related to the current load rate, and monitor the bus voltage drop recovery rate after each voltage reduction until the output current is lower than the cutoff threshold and then perform physical disconnection. The redistribution module is used to assess the phase space stability margin of the remaining healthy modules during the physical disconnection of the target module, and to distribute the missing power to each healthy module according to the phase space stability margin ratio. The collaborative early warning module is used to encapsulate the soft fault confidence and the selective cut-off instruction into a collaborative message and broadcast it to the distributed controller group to control adjacent modules to perform energy storage element pre-charging operation; The correction module is used to extract the normal vector offset of the phase space boundary surface based on historical load start feature data and erroneous cut-off records, and to reverse correct the divergence feature value judgment threshold in the reconstruction identification module.

2. The system according to claim 1, characterized in that, The process of extracting the envelope and separating the high-frequency disturbance of the output current transient sequence to generate the current base waveform and the high-frequency ripple echo attenuation factor includes: Extreme point detection is performed on the transient sequence of the output current, and the local maxima are connected to generate the upper envelope, and the local minima are connected to generate the lower envelope. The mean sequence of the upper and lower envelopes is calculated as the current base waveform; Extract the high-frequency components above a preset frequency band threshold from the transient output current sequence, calculate the amplitude attenuation ratio of the high-frequency components in adjacent zero-crossing intervals, and use the average of the amplitude attenuation ratios as the high-frequency ripple echo attenuation factor.

3. The system according to claim 1, characterized in that, The construction of the energy-impedance coupling spatial feature set includes: Extract the first phase angle at the moment of current change in the current base waveform, and the second phase angle at the moment when the voltage drops to the valley value in the bus voltage ripple sequence; Calculate the absolute value of the phase difference between the first phase angle and the second phase angle, and integrate the absolute value of the phase difference over time within a preset integration window to obtain the microscopic energy dissipation integral. The microscopic energy dissipation integral and the high-frequency ripple echo attenuation factor are orthogonally combined to form the energy-impedance coupling spatial feature set.

4. The system according to claim 1, characterized in that, The generation of soft fault confidence includes: A three-dimensional phase space is constructed using the dissipation integral of the energy-impedance coupling spatial feature set as the first coordinate axis, the echo attenuation factor as the second coordinate axis, and the time derivative as the third coordinate axis. In the three-dimensional phase space, calculate the curvature of the principal normal vector of the trajectory curve formed by the current working point and the previous preset number of working points, and denot it as the local manifold curvature; The variance change rate of the local manifold curvature within the sliding observation window is calculated as the phase space divergence characteristic value; When the phase space divergence characteristic value exceeds the reference divergence interval, the soft fault confidence is generated according to the excess ratio mapping.

5. The system according to claim 1, characterized in that, The generation of selective resection instructions includes: The soft fault confidence level is compared with a preset lower threshold and a higher threshold. The level below the lower threshold is the healthy range, the level above the higher threshold is the confirmed fault range, and the level in between is the fuzzy observation range. When the device is in the fuzzy observation range, the pulse width modulator of the corresponding power module is controlled to superimpose a voltage micropulse with a preset duty cycle width. The time constant required for the output current to fall back to the baseline after the voltage micro-pulse disappears is collected, and the reciprocal of the time constant is defined as the elastic recovery coefficient; When the elastic recovery coefficient is lower than the preset elastic lower limit, it is determined to be an internal soft fault, and the selective cut-off command is generated; otherwise, it is determined to be an external inductive load inrush, and the cut-off operation is shielded.

6. The system according to claim 1, characterized in that, In response to the selective disconnection command, the current load rate of the target module is extracted, the voltage setpoint of the target module is reduced in a step size inversely related to the current load rate, and the bus voltage drop recovery rate is monitored after each voltage reduction until the output current is lower than the cutoff threshold, at which point physical disconnection is performed, including: The selective cut-off command is parsed to obtain the target module identifier, and the ratio of the current output power to the rated power corresponding to the target module identifier is read as the current load rate; Set a step size for the derating rate, which is inversely proportional to the current load factor; The voltage setpoint of the target module is gradually reduced according to the step derating step size, and the bus voltage drop recovery rate within the set time window is calculated after each reduction. When the drop recovery rate is lower than the safe recovery baseline, pause the next long adjustment and maintain the current voltage setpoint until the recovery rate meets the requirements, then continue to adjust downwards. When the output current of the target module drops to the preset cutoff threshold, the physical contactor is controlled to disconnect, thus completing the physical disconnection.

7. The system according to claim 1, characterized in that, During the physical disengagement of the target module, the phase space stability margin of the remaining healthy modules is assessed, and the missing power is allocated to each healthy module according to the phase space stability margin ratio, including: Obtain the multidimensional distance from the current operating point of each health module to the preset stable attractor boundary in the three-dimensional phase space, and define the multidimensional distance as the phase space stability margin; The phase space stability margin of each health module is normalized and calculated to obtain the phase space stability margin ratio. Calculate the missing power carried by the physical disconnect module, allocate the missing power to the corresponding healthy module according to the phase space stability margin ratio, and generate the output current correction command for each healthy module.

8. The system according to claim 1, characterized in that, The step of encapsulating the soft fault confidence and the selective clearing command into a cooperative message and broadcasting it to the distributed controller group to control adjacent modules to perform energy storage element pre-charging operations includes: The module identifier of the soft fault, the confidence level of the soft fault, and the progress status of the step derating are encapsulated into the collaborative message. The cooperative message is broadcast to neighboring controllers in the distributed controller group via the internal bus; The adjacent controller parses the cooperative message, and when the soft fault confidence exceeds the warning benchmark, it calculates the pre-allocated power increment that its own module will bear. The pre-excitation current of the internal filter inductor or the pre-charge voltage of the capacitor are adjusted in advance according to the pre-allocated power increment, so that the energy storage element enters the energy pre-reservation state.

9. The system according to claim 1, characterized in that, The step of extracting the normal vector offset of the phase space boundary surface based on historical load start-up feature data and erroneous shearing records, and then reversing the divergence feature value judgment threshold in the reconstruction identification module, includes: Extract records of erroneous events from historical data that were initially identified as faulty zones but were ultimately confirmed as normal load startups; Extract the trajectory distribution of the working points recorded in the three-dimensional phase space of the malfunction event, and fit it to generate the actual boundary surface; Calculate the deviation of the normal vector angle between the actual boundary surface and the preset theoretical boundary surface at each sampling point, and use the mean of the normal vector angle deviation as the normal vector offset. The normal vector offset is multiplied by the damping attenuation coefficient, superimposed on the current divergence feature value determination threshold, and written into the reconstruction identification module.

10. The system according to claim 1, characterized in that, After the adjacent control module performs the pre-charging operation of the energy storage element, the method further includes: Monitor the overshoot value of the bus voltage at the moment the target module is physically disconnected; When the voltage overshoot amplitude exceeds the preset overshoot tolerance range, the ratio of the voltage overshoot amplitude to the upper limit of the overshoot tolerance range is calculated. The step size of the step-wise de-rating of the step-wise flexible disengagement module is reduced in the next de-rating operation according to the aforementioned excess ratio.

Images