Power distribution switch control device remote operation and maintenance management method and system

Abstract

The application relates to a power distribution switch control device remote operation and maintenance management method and system, and relates to the technical field of building power distribution equipment management. According to a sampling control instruction, an edge intelligent terminal synchronously provides time service and packet encapsulation for multi-source operation data of a power distribution switch according to a frequency, and the data is uploaded through a cellular network and an optical fiber Ethernet double-channel concurrent transmission; a cloud end estimates a link delay difference reference value online according to a difference between an absolute time stamp and an arrival time in a double-link packet stream and buffers the reference value as a disordered packet queue; based on combined braking, current mutation identification and twin forward state characteristic vector identification of an operation condition, a sliding window is amplified in a steady state and is forced to contract in a transient state; a target window parameter is cut off, dequeued and rearranged in ascending order, a missing period is physically dynamically interpolated in a phase space to obtain a time series reconstruction high-fidelity sequence; finally, a digital twin model is injected to perform virtual-real mapping deduction and mixed diagnosis, and a health degree score and a residual life prediction value are output.

Description

Technical Field

[0001] This application relates to the field of building power distribution equipment management technology, specifically to a remote operation and maintenance management method and system for power distribution switch control equipment. Background Technology

[0002] With the continuous advancement of intelligent transformation of power distribution networks, online monitoring and remote operation and maintenance of power distribution switch control equipment have become crucial links in ensuring power supply reliability. In existing technologies, edge intelligent terminals typically employ a dual-channel redundant architecture of cellular networks and power fiber optic Ethernet to transmit equipment operation data to the cloud platform, balancing communication reliability and real-time performance. However, due to factors such as air interface scheduling, base station load, and core network forwarding, cellular networks exhibit significant random jitter in transmission latency, creating a persistent asymmetric transmission latency difference compared to the fixed and extremely low latency of fiber optic channels. When two links concurrently transmit continuous sampling messages of the same physical process, millisecond-level message arrival out-of-order is inevitable at the cloud receiving side; that is, physically later event messages may arrive before preceding event messages. For the rapid action process of a power distribution switch, which takes only tens of milliseconds from coil excitation to contact closure, this timing reversal directly disrupts the rigid dependence of the state transition equation in the digital twin model on the causal timing of the input data, leading to divergence in the virtual-real mapping deduction or triggering false fault alarms.

[0003] To address the aforementioned out-of-order packet issue, existing solutions often borrow from the static jitter buffer strategy in the communications field. This involves setting a fixed-length waiting window in the cloud to passively collect out-of-order packets before reordering and injecting them into the model. While this approach can restore timing integrity to some extent, selecting the appropriate window size presents a dilemma: an excessively large window significantly increases end-to-end system latency, making it difficult to meet the stringent requirements of millisecond-level real-time diagnosis of distribution switches under transient conditions such as short-circuit fault clearing; an excessively small window cannot adequately accommodate the high latency jitter of cellular links, resulting in a large number of packets being lost beyond the window's capacity. This forces the downstream digital twin model to perform numerical integration interpolation over a significant time span, and the resulting cumulative truncation error severely impacts the accuracy of health scores and remaining lifetime predictions. Therefore, how to adaptively balance data integrity and inference fidelity without sacrificing real-time response capabilities is a pressing technical problem to be solved in the field of remote operation and maintenance management of distribution switches. Summary of the Invention

[0004] This application provides a method and system for remote operation and maintenance management of power distribution switch control equipment, so as to at least solve the above-mentioned technical problems existing in the prior art.

[0005] According to the first aspect of this application, a remote operation and maintenance management method for power distribution switch control equipment is provided, comprising: Step 1, using an edge intelligent terminal to synchronize and encapsulate the multi-source operation data of the power distribution switch according to the frequency assigned by the sampling and control command of the power distribution switch, and concurrently uploading it through a dual-channel cellular network and fiber optic Ethernet to obtain a dual-link message stream; Step 2, based on the difference between the absolute timestamp and the arrival time in the dual-link message stream, estimating the dual-channel transmission delay deviation online to obtain a link delay difference reference value, and buffering it into an out-of-order message queue according to the arrival order; Step 3, based on the closing / opening and current change identifiers and twin forward state feature vectors in the out-of-order message queue, identifying... The process involves identifying the operating conditions of the power distribution switch. In the steady state without operation, the sliding window is amplified using the link delay difference benchmark value. In the transient state with operation, the sliding window is forcibly shrunk using a preset limit truncation delay to obtain the target sliding window parameters and the physical interpolation enable signal. In the fourth step, the out-of-order message queue is truncated and dequeued according to the target sliding window parameters and rearranged in ascending order by timestamp. When the physical interpolation enable signal is valid, the missing time period messages are dynamically interpolated in physical phase space to obtain a high-fidelity time sequence reconstruction. In the fifth step, the high-fidelity time sequence reconstruction is injected into the digital twin model for virtual-real mapping deduction and hybrid diagnosis to output the power distribution switch health diagnosis results, which include health scores and remaining life prediction values.

[0006] According to a second aspect of this application, a remote operation and maintenance management system for power distribution switch control equipment is provided, comprising: an edge acquisition and encapsulation module, used to use an edge intelligent terminal to synchronize and encapsulate multi-source operating data of the power distribution switch according to the frequency assigned by the sampling and control command of the power distribution switch, and to concurrently upload the data through dual channels of cellular network and fiber optic Ethernet to obtain a dual-link message stream; a delay analysis and caching module, used to estimate the dual-channel transmission delay deviation online based on the difference between the absolute timestamp and the arrival time in the dual-link message stream to obtain a link delay difference reference value, and to cache the messages in the order of arrival as an out-of-order message queue; and a twin state perception module, used to detect the closing and opening circuit breakers and current change indicators in the out-of-order message queue and the twin forward state characteristics. The system comprises four modules: a eigenvector module and a hybrid diagnostic module. The eigenvector identifies the operating conditions of the distribution switch. In the steady-state condition without operation, the sliding window is amplified using the link delay difference benchmark. In the transient condition with operation, the sliding window is forcibly shrunk using a preset limit truncation delay, yielding the target sliding window parameters and a physical interpolation enable signal. The timing reconstruction and interpolation module truncates out-of-order message queues according to the target sliding window parameters and rearranges them in ascending order by timestamp. When the physical interpolation enable signal is valid, it performs dynamic physical phase space interpolation on messages with missing time periods to obtain a high-fidelity timing reconstruction sequence. The hybrid diagnostic module injects the high-fidelity timing reconstruction sequence into a digital twin model for virtual-real mapping and hybrid diagnostics, outputting a distribution switch health diagnostic result that includes a health score and remaining life prediction.

[0007] Compared with existing technologies, the remote operation and maintenance management method and system for power distribution switch control equipment provided in this application have the following technical effects: The remote operation and maintenance management system for power distribution switches achieves a mechanism leap from passive timeout eviction at the network layer to active coverage guidance at the physical layer at the data transmission level. During the steady-state monitoring period when the equipment is not in operation, the sliding window is automatically amplified based on the online estimated dual-channel delay difference, enabling late-arriving packets from high-jitter cellular links to be completely collected and rearranged according to absolute timestamps within a reasonable waiting time. This eliminates the causal inversion phenomenon caused by asymmetric delay of the link, ensuring that the data injected into the digital twin model has uniform and sufficient coverage characteristics in the physical time domain. When the distribution switch performs closing or opening operations or experiences short-circuit tripping or other transient sensitive conditions, the sliding window is forcibly shrunk to the limit truncation size. The system diagnostic response delay is reduced from hundreds of milliseconds in the traditional scheme to single-digit milliseconds, which is comparable to the fiber optic transmission delay. This meets the stringent real-time requirements of power grid fault clearing. For message periods lost due to window shrinkage, the forward interpolation of the physical mechanism differential equation limits the range of the digital twin numerical integral to a very small interval. This effectively suppresses the deteriorating effect of long-range integral cumulative truncation error on the accuracy of health score and remaining lifetime prediction. This enables the remote operation and maintenance system to achieve an adaptive dynamic balance among data integrity, inference fidelity, and diagnostic real-time performance. Attached Figure Description

[0008] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, wherein: in the drawings, the same or corresponding reference numerals denote the same or corresponding parts.

[0009] Figure 1 This is an overall flowchart of the remote operation and maintenance management method for power distribution switch control equipment according to an embodiment of this application; Figure 2 A schematic flowchart of step one in the remote operation and maintenance management method for power distribution switch control equipment according to an embodiment of this application is shown; Figure 3 A schematic flowchart of step two in the remote operation and maintenance management method for power distribution switch control equipment according to an embodiment of this application is shown; Figure 4 A schematic flowchart of step three in the remote operation and maintenance management method for power distribution switch control equipment according to an embodiment of this application is shown; Figure 5 A schematic flowchart of step four in the remote operation and maintenance management method for power distribution switch control equipment according to an embodiment of this application is shown; Figure 6 This is a schematic block diagram of a remote operation and maintenance management system for power distribution switch control equipment according to an embodiment of this application. Detailed Implementation

[0010] To further illustrate the technical means and effects adopted by this application in order to achieve the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of this application is provided in conjunction with the accompanying drawings and preferred embodiments.

[0011] Figure 1 This is an overall flowchart of the remote operation and maintenance management method for power distribution switch control equipment according to an embodiment of this application, such as... Figure 1 As shown, this application provides a method for remote operation and maintenance management of power distribution switch control equipment, including: Step one involves using an edge intelligent terminal to synchronize and encapsulate the multi-source operating data of the power distribution switch according to the frequency assigned by the sampling and control command. This data is then simultaneously uploaded via dual channels—cellular network and fiber optic Ethernet—to obtain a dual-link message stream. The multi-source operating data includes the current waveforms of the closing and opening coils, the displacement of the energy storage spring, the insulation resistance value, and the ambient temperature and humidity values. It is understandable that remote operation and maintenance management of power distribution switch control equipment relies on the continuous acquisition and reliable transmission of equipment operating status data. In actual operation, power distribution switches involve various physical processes such as electromagnetic transitions, mechanical actions, and insulation degradation. A single type of monitoring data cannot comprehensively reflect the overall health status of the equipment; therefore, it is necessary to simultaneously collect multi-source heterogeneous data, including electrical, mechanical, and environmental characteristics.

[0012] Meanwhile, the operation of power distribution switches is extremely rapid, with the time from coil energization to contact closure or separation typically only tens of milliseconds. This requires strict synchronization and alignment of various sensor data in the time dimension; otherwise, the cloud-based digital twin model will be unable to correctly establish the causal relationships between various physical quantities when performing virtual-real mapping simulations. Furthermore, a single communication link may experience data loss when facing abnormal conditions such as network congestion, link interruption, or base station switching. Adopting dual-channel redundant concurrent transmission of cellular networks and fiber optic Ethernet can effectively improve the reliability of data transmission.

[0013] Furthermore, the required monitoring accuracy for power distribution switches varies at different stages of degradation. When the equipment is in good health, a lower sampling frequency is sufficient for trend monitoring; however, when the equipment's health declines and the degradation trend accelerates, a higher sampling frequency is needed to capture more subtle degradation characteristics. Therefore, the sampling frequency of edge terminals should not be fixed but dynamically adjusted according to sampling control commands issued from the cloud to achieve on-demand allocation of sampling resources.

[0014] In one embodiment of this application, such as Figure 2As shown, step one includes: S11, based on the analog-to-digital conversion frequency of the dynamically modulated deterioration warning weight coefficient parsed from the power distribution switch sampling and control command, synchronous high-frequency sampling and discrete sequence splicing of the multi-source operation data of the power distribution switch are performed to obtain the digital feature tuple of the switch; S12, using the network path delay offset and local crystal oscillator drift error measured by the precise clock synchronization protocol, absolute time stamp calibration and frame header writing are performed frame by frame on the digital feature tuple of the switch to obtain the switch timing feature tuple; S13, after encapsulating the switch timing feature tuple with a unique device identifier, it is uploaded concurrently through dual channels of cellular network and fiber optic Ethernet to obtain a dual-link message stream.

[0015] First, the microprocessor of the edge intelligent terminal unpacks the received power distribution switch sampling and control command and extracts the degradation level warning weight coefficient from the command payload. Then, based on the pre-stored reference sampling frequency and this weight coefficient, the microprocessor calculates the target sampling frequency for the current service cycle using a dynamic sampling frequency modulation function. The modulation function is calculated as follows: the target sampling frequency equals the reference sampling frequency multiplied by a frequency amplification factor. The frequency amplification factor is controlled by an indicator function. When the degradation level warning weight coefficient is greater than a preset state degradation alarm threshold, the frequency amplification factor equals one plus a sampling amplification gain coefficient; when the weight coefficient does not exceed the alarm threshold, the frequency amplification factor is one, i.e., the reference frequency remains unchanged.

[0016] The above dynamic sampling frequency modulation function can be expressed as:

[0017] in, The target dynamic sampling frequency, This is the reference sampling frequency when the device is in normal operating condition. This is the sampling amplification gain coefficient. The degradation level warning weighting coefficient is obtained by parsing the control commands from the power distribution switch. The preset state degradation alarm threshold, This is an indicator function; it takes the value of one when the condition within the parentheses is true, and the value of zero otherwise.

[0018] After calculating the target sampling frequency, the microprocessor writes the frequency value into the control register of the analog-to-digital converter (ADC), driving the underlying hardware to synchronously sample the closing and opening coil current waveforms, energy storage spring displacement, insulation resistance, and ambient temperature and humidity values ​​of the power distribution switch at the new frequency. The ADCs of each channel share the same clock source trigger signal, ensuring strict alignment of the sampling times for different physical quantities. The sampled discrete-time series data are then concatenated in the terminal memory to generate a unified format of digital feature tuples for the switch.

[0019] Then, the edge intelligent terminal runs the IEEE 1588 precise time synchronization protocol, exchanging synchronization messages with the upstream master clock node to measure the network path delay offset between the local clock and the global master clock, as well as the inherent frequency drift error accumulated by the local crystal oscillator during operation. Using the above measurement results, the terminal calibrates and compensates the count value of the local hardware clock, and calculates the global absolute synchronization time corresponding to the current sampling point.

[0020] The calculation method for this time calibration is as follows:

[0021] in, For global absolute time synchronization, Let be the local hardware clock count value of the edge intelligent terminal at the k-th sampling instant. This refers to the static delay offset of the one-way network path between the master and slave devices, measured using a precise time synchronization protocol. This represents the nonlinear intrinsic frequency drift error function generated by the local crystal oscillator over time. This is the time when the clock synchronization calibration was last completed. This represents the time when the current k-th sampling occurs.

[0022] After completing the calibration calculation, the terminal traverses each data frame in the digital feature tuple of the switch and writes the corresponding global absolute synchronization time as a high-precision timestamp into the frame header structure to generate a switch timing feature tuple carrying an absolute time stamp.

[0023] Finally, the edge intelligent terminal invokes a pre-built industrial IoT serialization encoding algorithm to perform low-level bytecode encoding on the switch timing feature tuple. The edge intelligent terminal's unique device identifier and an incrementing message sequence number are appended to the header of the encoded byte stream, completely encapsulating it into a standard transmission payload. Subsequently, the terminal performs a deep copy operation on the standard transmission payload in memory, generating two identical payload copies. The terminal wakes up its internally independent cellular network communication module and fiber optic Ethernet communication module, pushing the two payload copies into their respective transmission buffers and sending them out in parallel, thus forming a redundant dual-link message stream at the physical link layer.

[0024] In one specific embodiment, the edge intelligent terminal is deployed in an intelligent terminal box on the side of the 10kV distribution switch cabinet, establishing a data channel with the distribution switch control equipment via an RS485 interface and a high-speed analog signal acquisition board. The reference sampling frequency is set to 10kHz, which meets the basic feature extraction requirements of the coil current waveform under normal monitoring conditions. The sampling amplification gain coefficient is set to 1.5, meaning that when high-frequency sampling is triggered, the target sampling frequency increases from 10kHz to 25kHz. The status degradation alarm threshold is set to 0.6, meaning that when the degradation level warning weight coefficient issued by the cloud exceeds 0.6, it indicates that the equipment has entered a critical period of accelerated degradation requiring close monitoring, and the terminal automatically switches to high-frequency sampling mode.

[0025] Regarding time synchronization, the edge terminal's built-in precise time synchronization protocol module exchanges synchronization messages with the substation's master clock server every second. In the actual test environment, the typical value of the network path static delay offset is between 0.2 milliseconds and 2 milliseconds, and the local crystal oscillator drift rate is approximately ±20 ppm, corresponding to a cumulative drift of approximately ±20 microseconds within one synchronization cycle (1 second). After calibration and compensation, the absolute timestamp accuracy of each data frame can reach the microsecond level, which can meet the accuracy requirements for causal timing calibration of various physical quantities during the millisecond-level operation of the power distribution switch.

[0026] Regarding the communication links, the cellular network module uses a 4GLTECat.4 module with an uplink bandwidth of 50Mbps and a typical transmission latency of 30 to 80 milliseconds, which can reach 150 milliseconds during base station handover or network congestion. The fiber optic Ethernet module is accessed through a dedicated power fiber optic ring network, with a transmission rate of 100Mbps and a typical transmission latency of less than 5 milliseconds with minimal fluctuation. Both modules share the same encoded message payload and independently complete link-layer encapsulation and physical-layer transmission, ensuring that the other link can still independently complete the data upload task in the event of a failure in either link.

[0027] Step two: Based on the difference between the absolute timestamp and the arrival time in the dual-link message stream, the dual-channel transmission delay deviation is estimated online to obtain the link delay difference benchmark value, and the messages are cached as an out-of-order queue according to their arrival order. It should be understood that when an edge intelligent terminal concurrently uploads power distribution switch operation data through dual channels of 4G / 5G cellular networks and power line fiber optic Ethernet, the two heterogeneous links have fundamental differences in their physical transmission mechanisms. Power line fiber optic Ethernet has a fixed path, extremely low transmission delay, and strong determinism, while the cellular network is affected by factors such as air interface scheduling, base station load, and core network gateway forwarding, resulting in a significant random jitter distribution in transmission delay. This asymmetric transmission delay characteristic leads to a discrepancy between the arrival order of continuously sampled messages for the same physical process at the cloud receiving side and their actual sampling order at the edge side, i.e., a message timing reversal phenomenon.

[0028] To correctly reorder out-of-order packets, the cloud-based operations and maintenance management platform needs to monitor the magnitude and dynamic trend of the transmission delay deviation between the two links in real time. This delay deviation determines how long the subsequent sliding window should wait before reasonably expecting the arrival of late packets, and is the core input basis for downstream adaptive sliding window scaling decisions. Simultaneously, since packets enter the system in network arrival order rather than physical sampling order, they need to be cached in a queue structure for subsequent truncation and reordering processing.

[0029] In one embodiment of this application, such as Figure 3 As shown, step two includes: S21, extracting physical payload and projecting arrival time attributes on the dual-link message stream to obtain a dual-timestamped message sequence; S22, performing online link delay jitter estimation on the dual-timestamped message sequence to obtain a link delay difference benchmark value and a set of timestamped messages; S23, pushing messages into a first-in-first-out circular buffer in ascending order of arrival time, physically caching and logically out-of-order instantiation of the timestamped message set to obtain an out-of-order message queue.

[0030] First, the access gateway of the cloud-based operations and maintenance management platform continuously listens for and receives dual-link message streams from the edge side. The gateway deserializes and unpacks the underlying protocol streams, extracting the physical payload of the power distribution switch operation data and the absolute timestamp already bound on the edge side. The moment a message arrives at the cloud gateway, the gateway invokes the high-precision clock of the cloud system kernel to add the cloud's local arrival time to each successfully unpacked message. Subsequently, the original payload, the edge-side absolute timestamp, the cloud arrival time, and the link type identifier are combined to generate a structured, dual-timestamped message sequence.

[0031] The process of combining these attributes can be expressed as follows:

[0032] in, For the k-th double-time-stamped message tuple, To extract the native digital load data of the physical operation of the distribution switch, To parse the absolute timestamp assigned by the edge side from the message header, This is the arrival time written by the kernel the instant the message arrives at the cloud gateway. The physical link type identifier for carrying this message transmission, with a value of either cellular network or fiber optic Ethernet.

[0033] Then, the cloud system traverses the dual-timestamped message sequence and, based on the device unique identifier and message sequence number in the message header, matches and reassembles redundant message pairs from the same physical event originating from cellular network links and fiber optic links. For successfully matched message pairs, the absolute deviation between their arrival times in the cloud is calculated as the instantaneous network transmission delay difference for the current sampling period.

[0034] The instantaneous delay difference is calculated as follows: the instantaneous network arrival delay difference of the k-th paired message. Equal to the time it takes for the same physical event message to reach the cloud via the cellular link. The time it takes to reach the cloud via fiber optic link The absolute value of the difference, that is:

[0035] Because cellular network latency exhibits high-frequency random jitter, a single instantaneous latency difference is insufficient to represent the overall network state. Therefore, an exponential moving average algorithm is used to apply a first-order smoothing filter to the instantaneous latency difference, calculating the smoothed expected value of the latency difference. Simultaneously, a second-order exponential smoothing is used to calculate the dynamic variance of the latency difference to capture the severity of network jitter. Finally, the smoothed mean is added to an additional variance margin to serve as the reference value for the link latency difference used as the basis for the out-of-order window.

[0036] The specific formulas for the above calculation process are as follows:

[0037]

[0038]

[0039] in, This represents the first-order smoothed expected value of the network delay difference at the current sampling point. This is the smoothed expected value of the previous sampling point. It is the time decay factor of the first-order exponential moving average, which controls the sensitivity of the mean to recent network fluctuations. This represents the second-order dynamic variance of the network delay difference fluctuation during the current period. This is the update weight constant for variance smoothing filtering. This is the final output reference value for the link delay difference. This is the fault tolerance margin coefficient for network fluctuations, used to add sufficient safety margin to the smoothed mean to cover the tail distribution of network jitter.

[0040] Finally, the cloud system allocates a contiguous first-in-first-out circular buffer area in the server memory. Each packet element with extracted delay information is pushed into the buffer area one by one strictly in accordance with the ascending order rule based on its cloud arrival time. Due to the objective existence of the delay difference of heterogeneous networks, the physical queue sorted by the cloud arrival time will logically necessarily break the causal sequence reflected by the edge-side absolute timestamp natively embedded in the packets, thereby forming an out-of-order packet queue that is physically arranged in sequence but logically misaligned in time sequence.

[0041] The formal constraints of this queue can be expressed as: the elements in the out-of-order packet queue satisfy that for any position index i<j, there must be an arrival time , but the absolute timestamp and do not maintain an equivalent monotonically increasing relationship, that is:

[0042] Herein, is the instantiated out-of-order packet queue, and are any two packet elements arranged by index in the queue, is the full set of packets before being pushed into the queue, is the cloud arrival time attribute of the corresponding packet, is the edge-side absolute timestamp attribute of the corresponding packet, indicates that the equivalent monotonicity is lost between the two time attribute sequences.

[0043] In a specific embodiment, the cloud operation and maintenance management platform is deployed in the server cluster of the power dispatching data center, and the access gateway is configured with dual network cards to respectively connect to the 4G dedicated APN channel and the power fiber Ethernet OTN channel.

[0044] For the value of the delay attenuation factor, it is set to 0.15 in actual deployment. This value means that the contribution weight of the instantaneous delay difference observed most recently to the smoothed mean is 15%, and the contribution weight of the historical cumulative mean is 85%. This setting enables the smoothed mean to track the medium-term change trend of the network delay, while not being excessively interfered by occasional extreme jitters.

[0045] For the value of the variance update weight, it is set to 0.10. This value makes the update speed of the dynamic variance slightly slower than the mean update, preventing the variance estimation from drastic jump due to individual abnormal values, while still being able to reflect the substantial change of the network jitter intensity within more than ten sampling cycles.

[0046] The fault tolerance margin coefficient is set to 2.5. This value is based on the confidence interval of the Gaussian distribution, adding 2.5 times the standard deviation to the smoothed mean as a safety margin, which can cover about 98.8% of normal network jitter distribution. Only a very small number of extremely late packets that exceed this margin will be truncated by the subsequent window.

[0047] In real-world operating environments, the typical transmission latency of power line fiber optic Ethernet is 2 to 5 milliseconds with fluctuations within 1 millisecond, while the typical transmission latency of 4G cellular networks is 30 to 80 milliseconds, reaching up to 150 milliseconds during base station handover or sudden increases in network load. Through continuous calculations using the aforementioned online estimation algorithm, the baseline value for link latency difference typically stabilizes between 50 and 90 milliseconds during periods of network stability, and can be dynamically adjusted to between 120 and 160 milliseconds during periods of increased network volatility. This dynamic adjustment feature allows subsequent sliding windows to be set with appropriate waiting times based on real-time network conditions, avoiding the dilemma of fixed windows either wasting real-time performance due to excessively long waiting times or losing critical data due to excessively short waiting times when network conditions change.

[0048] The capacity of the circular buffer is configured by dividing the maximum allowed buffer duration by the minimum sampling interval. With a base sampling frequency of 10kHz and a maximum buffer duration of 500ms, the circular buffer needs to hold approximately 5000 packet frames. When the buffer is full, the earliest arriving packet is automatically overwritten to prevent memory overflow, and the system triggers a timeout alarm to notify downstream modules of potential data loss.

[0049] Step 3: Based on the closing / opening and current surge identifiers in the out-of-order message queue and the twin forward state feature vector, the operating conditions of the distribution switch are identified. In the steady-state without operation, the sliding window is enlarged using the link delay difference reference value; in the transient operation, the sliding window is forcibly contracted using a preset limit truncation delay, thus obtaining the target sliding window parameters and the physical interpolation enable signal. It should be understood that the size of the sliding window cannot simply be equal to and remain unchanged by the link delay difference reference value. This is because the distribution switch control equipment experiences two drastically different operating states in actual operation, with vastly different real-time requirements for data processing.

[0050] When the distribution switch is in a steady-state without operation, the equipment only needs to perform routine health trend monitoring and has a high tolerance for diagnostic delays. In this case, the sliding window should be enlarged as much as possible to fully wait for late messages on the cellular network link, ensuring data integrity and coverage. When the distribution switch is in transient conditions such as closing / opening or tripping due to short-circuit faults, the entire physical process from coil excitation to complete contact separation or closure lasts only tens of milliseconds. The cloud-based digital twin model needs to complete state deduction and fault diagnosis in a very short time to support power grid protection decisions. If a large sliding window is maintained to wait for late messages in this situation, the diagnostic response delay will increase by tens or even hundreds of milliseconds, which cannot meet the stringent requirements of millisecond-level real-time diagnosis under transient conditions.

[0051] Therefore, a mechanism is needed that can sense the current operating condition of the power distribution switch in real time and dynamically adjust the sliding window size accordingly. This mechanism integrates abrupt change characteristic information from the actual physical side with forward state inference information from the digital twin virtual side to comprehensively determine the current operating condition of the equipment, and adaptively expands or contracts the sliding window based on the determination result, while generating an enable signal to control whether the downstream initiates physical interpolation.

[0052] In one embodiment of this application, such as Figure 4 As shown, step three includes: S31, performing lightweight window scanning and mutation feature extraction on the out-of-order message queue to obtain the switch local mutation feature vector; S32, performing virtual and real state joint splicing and transient probability inference on the switch local mutation feature vector and twin forward state feature vector to obtain the operating condition transient probability score; S33, based on the decision result between the operating condition transient probability score and the preset service sensitivity threshold, performing window scaling and interpolation enable setting on the operating condition transient probability score and the link delay difference benchmark value to obtain the target sliding window parameters and physical interpolation enable signal.

[0053] First, the cloud system initiates a lightweight sliding observation window on the out-of-order message queue, reading only the most recently arrived observation messages from the head to the tail of the queue. The coil current and mechanical displacement values ​​in these messages are analyzed, and their local rates of change with time are calculated using a first-order difference algorithm. The absolute maximum value of the rate of change within the time window is extracted to construct a local abrupt change feature vector characterizing the intensity of fluctuations in the current physical state of the equipment.

[0054] The feature extraction is calculated as follows: local mutation feature vector It consists of the absolute maximum value of the first-order time rate of change of the coil current within the window and the absolute maximum value of the first-order time rate of change of the mechanical displacement, that is:

[0055] in, To calculate the output local mutation feature vector, Let k be the set of sliding observation windows selected from the queue, containing the latest N messages, where k is the index number of the discrete messages in the window set. This represents the first-order time rate of change of the coil current between adjacent messages within the window. This represents the first-order time rate of change of mechanical displacement between adjacent messages within the window.

[0056] It should be noted that the twin forward state feature vector is generated by the forward inference step of the digital twin model in step five and fed back to step three for condition identification in the next operation and maintenance cycle. When the system is first powered on or the equipment is first connected to the platform, there are no historical inference results available for reference. At this time, the twin forward state feature vector is initialized to a zero vector, that is, all feature dimensions are set to zero. Zero vector input means that condition identification relies only on the local mutation feature vector of the actual physical side for judgment. After the system has run for several cycles and accumulated effective twin inference results, virtual side information gradually participates in joint judgment, and the classification accuracy gradually improves with the increase of the operation cycle.

[0057] Secondly, the local mutation feature vector representing the actual physical mutation status is concatenated with the twin forward state feature vector representing the predicted trend of the cloud virtual model to complete the joint representation of the virtual and real states. The combined multidimensional vector is then input into a pre-trained logistic regression condition classifier. The classifier calculates a scalar value normalized to the interval between zero and one using the Sigmoid activation function, which is used to characterize the probability that the distribution switch is currently in a transient condition such as operation opening / closing or fault tripping.

[0058] The probability of this transient operating condition is calculated as follows: transient probability fraction of operating condition. This is equal to the mapping result of the Sigmoid function to the sum of the weighted values ​​of the physical mutation features, the weighted values ​​of the twin forward states, and the bias term, i.e.:

[0059] in, This represents the transient probability score of the operating condition output by the classifier, with a value ranging from zero to one. The local mutation feature vector obtained from the above calculation is... The input is the twin forward state feature vector. This is the learned weight matrix assigned to the physical mutation feature vector in the classifier. This is the learned weight matrix assigned to the twin forward state feature vectors in the classifier. This is the bias term constant of the classifier.

[0060] Finally, the transient probability score is compared with the system's preset service sensitivity threshold for a decision. If the probability score is lower than the service sensitivity threshold, the distribution switch is determined to be in a non-operational steady state. The decision logic sets the target sliding window parameter to be equal to the link delay difference benchmark value to fully accommodate network out-of-order data and ensure data integrity, while simultaneously disabling the physical interpolation enable signal. If the probability score is greater than or equal to the service sensitivity threshold, the distribution switch is determined to be in an agile transient state. The decision logic forcibly discards the link delay difference benchmark value, sets the target sliding window parameter to a preset limit truncation delay to ensure millisecond-level diagnostic response, and simultaneously enables the physical interpolation enable signal.

[0061] The calculation method for this dual-element control decision is as follows:

[0062]

[0063] in, The final target sliding window parameters, This is the physical interpolation enable signal. This is the baseline value for link delay difference. The minimum truncation delay preset by the system, The transient probability fraction of the operating condition. The system is preset with business sensitivity thresholds. This is an indicator function; it takes the value of one when the condition inside the parentheses is true, and the value of zero when the condition is false.

[0064] In one specific embodiment, the number of messages N in the lightweight sliding observation window is set to 20. Calculated at a reference sampling frequency of 10kHz, the physical time span covered by 20 messages is 2 milliseconds, which can capture the millisecond-level step characteristics of the current in the distribution switch coil at the moment of operation, without introducing excessive computational delay due to an excessively large window.

[0065] Weight matrix of logistic regression working condition classifier and The data was obtained through offline training. The training data came from a large number of closing and opening operations and simulated fault experiments performed on power distribution switches in a laboratory environment. After training, the classifier's parameters were stored in a cloud server. During online inference, only one matrix multiplication and sigmoid operation are required, and the time for a single inference is less than 0.1 milliseconds.

[0066] The business sensitivity threshold is set to 0.65. This value was determined through online verification and optimization. When the transient probability score exceeds 0.65, it indicates that the classifier has a high confidence that the device is undergoing an operation. If the threshold is too low, small current fluctuations during normal operation may be misjudged as transients, leading to frequent window contraction. If the threshold is too high, it may delay the response in the early stages of the actual operation.

[0067] The preset truncation delay is set to 5 milliseconds. This value is comparable to the typical transmission delay of power line fiber optic Ethernet, meaning that under transient conditions, the system only waits for the normal transmission time of fiber optic link messages before truncation, abandoning the waiting for late messages on the cellular link. For cellular link messages lost due to window shrinkage, forward interpolation using the physical mechanism differential equations in step four is used to compensate.

[0068] In actual operation, when the distribution switch receives a tripping command and begins to operate, the coil current rises from zero to a peak value of several amperes within 1 to 2 milliseconds, corresponding to a first-order time change rate of several thousand amperes per second. At this time, the current change rate component in the local mutation feature vector experiences an order-of-magnitude jump, which, combined with the predictive information in the twin forward state feature vector indicating that the equipment is about to operate, causes the transient probability score of the operating condition to rapidly exceed the service sensitivity threshold of 0.65. The sliding window is then forcibly contracted from the link delay difference reference value (typically 60 to 100 milliseconds) to 5 milliseconds, and the physical interpolation enable signal is synchronously set to the active state. When the distribution switch completes its operation, the coil current decays back to zero, and the mechanical displacement stabilizes, the local mutation feature vector returns to a low amplitude level, the transient probability score drops back below the threshold, the window automatically returns to the reference value size, and the enable signal is reset to the inactive state.

[0069] Step four involves truncating and dequeuing out-of-order messages according to the target sliding window parameters and rearranging them in ascending order of timestamps. Furthermore, when the physical interpolation enable signal is active, physical phase space dynamic interpolation is performed on messages in the missing time period to obtain a high-fidelity time sequence reconstruction. It should be understood that the messages in the out-of-order message queue are still physically arranged according to the order of their arrival time in the cloud, and their logical timing is inconsistent with the causal timing of the actual physical actions of the power distribution switch. The state transition equations and electro-magnetic-mechanical coupling differential equations in the cloud-based digital twin model have a rigid dependence on the causal timing of the input data. If the mechanical linkage displacement data is received before the coil trigger current data, the model's state transition matrix will not be correctly solved or will produce numerical abrupt changes, leading to a break in the deduction link.

[0070] Therefore, before injecting messages into the digital twin engine, messages that have reached their waiting period are truncated and dequeued according to the target sliding window parameters, and then strictly rearranged in ascending order according to their embedded edge-side absolute timestamps to restore the causal timing of physical actions. Furthermore, when the power distribution switch is in a transient state and the sliding window is forcibly contracted, some high-latency messages transmitted via the cellular network will inevitably fail to reach the cloud within the extremely short window time, resulting in physical time domain coverage gaps in the timing sequence. If an incomplete sequence containing these gaps is directly injected into the digital twin model, the model will have to perform numerical integration over a large time span, and the resulting cumulative truncation error will affect the accuracy of health scores and remaining life predictions. Therefore, when the physical interpolation enable signal is valid, the physical mechanism differential equations of the power distribution switch itself need to be invoked to perform forward interpolation on the missing time periods to bridge the gaps, ultimately outputting a high-fidelity feature sequence with complete causal timing and uninterrupted physical time domain.

[0071] In one embodiment of this application, such as Figure 5 As shown, step four includes: S41, comparing the difference between the current system time and the message arrival time with the target sliding window parameters item by item, performing timeout judgment and forced dequeueing truncation on the out-of-order message queue to obtain a truncated time-limited message set; S42, when the physical interpolation enable signal is valid, performing forward interpolation of missing time period states and virtual message union fusion on the truncated time-limited message set to obtain an interpolated fused extended message set; S43, using the edge side absolute timestamp as the unique sorting key, performing causal temporal rearrangement and linear continuous solidification on the interpolated fused extended message set to obtain a high-fidelity temporal reconstruction sequence.

[0072] First, obtain the current system absolute clock time of the cloud gateway, traverse all packets in the out-of-order packet queue, calculate the time difference between the current system time and the arrival time of each packet in the cloud, and determine that the packets with a time difference greater than or equal to the target sliding window parameter have reached the maximum waiting period and forcibly dequeue them from the queue, forming a set of truncated discrete packets.

[0073] In the first embodiment, the truncation decision logic is entirely anchored to the network layer attribute of the message's arrival time in the cloud. Specifically, the time difference between the current time of the cloud system and the arrival time of each message is calculated one by one. Once the time difference is greater than or equal to the waiting tolerance threshold set by the target sliding window parameter, the message is determined to have reached its maximum waiting period and is forcibly dequeued, ultimately forming a truncated discrete set. The formal constraint of this process is: the elements in the truncated time limit message set \mathbb{S}_{trunc} satisfy that the difference between the current system time and the message arrival time is not less than the target sliding window parameter, that is:

[0074] in, To truncate the time-limited message set, For the i-th discrete message element in the queue, For out-of-order message queues, This is the current absolute clock time when the cloud system gateway performs the check operation. For message Recorded arrival time in the cloud. Parameters for the target sliding window.

[0075] However, each packet in the out-of-order packet queue carries not only the network layer arrival time attribute, but also a crucial physical layer attribute: the edge-side absolute timestamp. This timestamp reflects the actual time when the physical action of the corresponding power distribution switch occurs, rather than the time it arrives at the cloud after being transmitted through the network. The truncation determination logic in the first embodiment completely fails to incorporate the coverage density distribution characteristics of the packet's edge-side absolute timestamp in the physical time domain into the decision-making basis.

[0076] This deficiency can lead to the following consequences in practical applications of remote operation and maintenance of power distribution switches. Due to the influence of air interface scheduling, base station load, and core network gateway forwarding, 4G / 5G cellular networks exhibit random jitter distributions of transmission latency ranging from 20 milliseconds to 150 milliseconds, while power fiber optic Ethernet paths are fixed and have extremely low latency. Under a truncation strategy with dual-channel concurrent upload and a unified timeout threshold, a large number of packets arriving rapidly via fiber optics are concentrated in the truncation set. The absolute timestamps on the edge side corresponding to these packets tend to cover certain specific time segments during the physical operation, causing local overcrowding. Simultaneously, packets transmitted slowly via cellular links have not yet timed out, and their corresponding physical time segments are completely missing from the truncation set, forming coverage holes.

[0077] Taking a complete tripping operation of a distribution switch as an example, the entire physical process from coil de-energization to complete contact separation lasts approximately 30 to 80 milliseconds. This process includes multiple strictly timed physical steps such as coil electromagnetic transition, energy storage spring release, and mechanical linkage movement. When the messages in the truncated set exhibit a non-uniform, alternating distribution in the physical time domain, dense sections contain redundant data, wasting storage and sorting computational power, while hollow sections completely lack actual measurement support. Since the downstream needs to inject the truncated data into the digital twin engine for virtual-real mapping simulation, the numerical solver must be activated to call the electro-magnetic-mechanical coupled differential equation system for forward integration interpolation for hollow sections. The larger the span of the physical time domain hollow, the more the numerical truncation and rounding errors in the forward integration process of the differential equations will accumulate and amplify, directly reducing the fidelity of the digital twin virtual-real mapping, and thus affecting the accuracy of the health score and remaining service life prediction output by the hybrid diagnostic engine.

[0078] To address the aforementioned technical deficiencies, in the second embodiment, the truncation determination logic is upgraded from a pure network layer timeout eviction to an adaptive layered truncation and targeted supplementary sampling mechanism that integrates physical temporal coverage density awareness. In this second embodiment, based on a step-by-step comparison of the difference between the system's current time and the arrival time of each message with the target sliding window parameters, the out-of-order message queue is subjected to timeout determination and forced dequeue truncation to obtain a truncation-limited message set. This includes: based on a step-by-step comparison of the difference between the system's current time and the arrival time of each message with the target sliding window parameters, the out-of-order message queue is subjected to timeout eviction and in-transit retention binary division to obtain a preliminary truncation message set and a queue residual candidate message set; temporal coverage density calculation and low-density hole identification are performed on the preliminary truncation message set to obtain a temporal coverage hole interval set; based on the temporal coverage hole interval set, the queue residual candidate message set is subjected to timestamp-targeted retrieval and early dequeue fusion to obtain a truncation-limited message set.

[0079] First, in the data reception stage of remote operation and maintenance of power distribution switches, the cloud gateway continuously receives the message stream uploaded concurrently from the edge terminal via dual channels and caches it in an out-of-order message queue. When a queue truncation operation is required, a basic timeout determination is first performed, and messages that have reached the maximum waiting tolerance period are removed from the queue. Regardless of subsequent physical time domain coverage analysis, messages whose waiting time has exceeded the window threshold are no longer suitable to remain in the buffer queue at the network communication semantic level. Continuous retention would lead to an infinite expansion of queue memory usage, and the physical state data carried by these messages already possesses a ready state for downstream consumption.

[0080] Specifically, the system obtains the current absolute clock time of the cloud gateway, traverses all packets in the out-of-order packet queue, calculates the time difference between the current system time and the arrival time of each packet in the cloud, and forcibly dequeues packets with a time difference greater than or equal to the target sliding window parameter to form an initial truncated packet set. Simultaneously, packets that have not yet timed out but have entered the queue are retained and explicitly marked as the queue's residual candidate packet set. By dividing the queue in two instead of simply extracting only expired packets and discarding the queue context, subsequent physical temporal coverage analysis can simultaneously grasp two clear operational objects: the truncated data pool and the remaining candidate data pool. This avoids the computational overhead of retracing the entire original queue during subsequent targeted re-sampling, while ensuring the traceability and retrieval of candidate packets.

[0081] Secondly, after global timeout truncation, the distribution characteristics of the packets in the initial truncated packet set in the physical time domain are still unclear. The absolute timestamps of the edge sides of all packets in the initial truncated packet set are extracted to determine the upper and lower bounds of their physical time span. This span is then divided into several micro-intervals. The number of packets falling into each micro-interval is counted, and this number is divided by the system's preset minimum fidelity density threshold for each micro-interval to obtain the physical time domain coverage density vector.

[0082] The calculation process for cover density is as follows:

[0083] in, Let be the physical temporal coverage density ratio of the j-th micro-interval. To initially screen and truncate the i-th discrete message element in the message set, The edge-side absolute timestamp carried in the message. The minimum absolute timestamp of all messages in the initial screening truncation set is the lower bound of the physical time span. This represents the time width of each micro-interval after the physical time span has been uniformly divided. The minimum message count threshold that each micro-interval should cover as preset by the system is the fidelity density benchmark.

[0084] Micro-intervals with a coverage density below one are designated as coverage void intervals. The set of void intervals is defined as follows:

[0085] in, This is a set of indexes for hollow micro-intervals with insufficient coverage density. This represents the total number of micro-intervals into which the physical time span is divided. When This means that the number of measured messages within this micro-interval is insufficient to support a reliable virtual-real mapping of the digital twin model during that period, requiring supplementary observation data from external sources or subsequent numerical compensation via differential equations.

[0086] Subsequently, the remaining candidate message set in the scan queue is retrieved, and messages whose absolute timestamps on the edge side fall exactly within the marked gap interval are extracted in advance and merged with the initially screened truncated message set. The set operation of targeted supplementary sampling fusion is expressed as:

[0087] in, This is the final set of truncated time-limited messages to be output. This is for the initial screening and truncation of the message set. This is the set of candidate messages remaining in the queue. By targeting and supplementing the data, data that can fill physical gaps is accurately extracted from slow-link messages that have not yet timed out. This avoids the downstream twin model from starting a full-span numerical integration operation for a small time-domain gap, thus compressing the growth space of the cumulative truncation error of numerical integration from the source.

[0088] Then, after targeted supplementary sampling and fusion, the physical temporal coverage integrity of the truncated time-limited message set has been improved compared to the initial screening truncated set. However, since there may not always be messages in the queue of remaining candidate messages that can accurately fall into the void interval, void intervals that cannot be resolved may still remain after supplementary sampling. Therefore, it is necessary to perform a secondary verification of the physical temporal coverage density on the supplemented truncated time-limited message set to dynamically determine whether downstream Runge-Kutta numerical integration still needs to be initiated.

[0089] Based on the aforementioned targeted supplementary sampling fusion, as a further preferred embodiment, the physical interpolation enable signal can be subjected to secondary verification and dynamic degradation. Specifically, the coverage density of the truncated time-limited message set is recalculated within micro-intervals to obtain the set of residual void micro-intervals after supplementary sampling. If the coverage density of all micro-intervals meets the minimum fidelity threshold requirement, i.e., the set of residual voids is empty, the physical forward interpolation enable signal is downgraded to an invalid state, indicating that there are no missing segments in the physical time domain, and there is no need to start differential equation forward interpolation again; if there are still unresolved void intervals after supplementary sampling, the original valid state of the physical forward interpolation enable signal remains unchanged. The logical expression of the joint determination of the enable signal is as follows:

[0090] in, This is the updated physical forward interpolation enable signal output after joint determination. This is the Boolean value of the original physical forward interpolation enable signal transmitted from upstream. This is the set of residual void micro-intervals obtained after re-verification following targeted re-mining. An empty set indicates that all holes have been successfully filled. This is an indicator function; it takes one if the condition within the parentheses is true, and zero otherwise. The time-indicative function output is zero, regardless of the original enable signal value, thus directly skipping the entire numerical integration stage of the downstream differential equation.

[0091] Secondly, the cloud-based time series reconstruction module checks the Boolean state of the physical interpolation enable signal. If the signal is invalid, the interpolation operation is skipped, and the truncated time-limited message set is directly used as the output of this sub-step. If the signal is valid, it indicates that due to the forced shrinking of the sliding window, there is a time-domain discontinuity in the time series caused by the loss of high-latency link data.

[0092] At this point, the system extracts the legitimate message with the timestamp closest to the fault front from the truncation time limit message set, and reads the coil current value, energy storage spring displacement value, and mechanical link velocity value carried in the message as the initial state boundary conditions. Subsequently, it calls the electro-magnetic-mechanical coupled differential equation set of the power distribution switch control equipment. This equation set covers the coil electromagnetic transient equation and the spring mechanical dynamics equation, describing the complete physical causal chain of the power distribution switch from control voltage establishment, coil current generation, magnetic circuit engagement to energy storage spring release and mechanical link movement.

[0093] The system applies a fourth-order Runge-Kutta numerical algorithm to perform continuous-time integration on the missing timestamp intervals, predicting the current and displacement state quantities of the equipment during the lost time period. The calculation method for this forward prediction is as follows:

[0094] in, This is a virtual physical state vector predicted in the continuous time domain t, containing state variables such as current, displacement, and velocity. This refers to the initial state vector of the known boundary at the fault front, extracted from the truncated time-limited message set. The electromagnetic-mechanical coupled differential equations, i.e., the state transition function, characterize the nonlinear characteristics of the power distribution switch control equipment. This is the time integration variable.

[0095] The predicted state variables are discretized according to the baseline sampling frequency to generate a set of virtual messages carrying the corresponding missing timestamps. This set is then joined with the truncated time-limited message set to obtain a spatially complete and expanded dataset. The virtual message set and the fusion operation are expressed as follows:

[0096]

[0097] in, This is a set of virtual messages generated after discretizing and sampling a continuous predicted trajectory. This refers to the open interval of the missing time period exposed by link out-of-order delivery and window shrinkage. To interpolate and merge extended message sets, The physical interpolation enable signal is used as the control operator to enable fusion through scalar multiplication.

[0098] Finally, the absolute timestamp on the edge of the header of each message (whether real or virtual) in the interpolated and fused extended message set is extracted. Using this absolute timestamp as the unique comparison key, a quicksort or mergesort algorithm is called to strictly rearrange all messages in the set in ascending order. During the sorting process, the causal inversion caused by the sequential arrival of fiber optic and cellular data due to heterogeneous network link switching is completely eliminated. The sorted set is converted into a linearly continuous array structure and solidified into a high-fidelity time-series reconstruction sequence with strong causal logic and no data gaps.

[0099] The logical expression of the refactoring process is as follows:

[0100] The rearranged sequence satisfies the strict monotonicity constraint:

[0101] in, To reconstruct a high-fidelity sequence for the final solidified output timing. This is an ascending order reordering algorithm that is executed with reference to a specified attribute. This is the absolute timestamp attribute on the edge side of the message, which serves as the sorting primary key.

[0102] In one specific embodiment, the power distribution switch performs a tripping operation, the physical action of which lasts approximately 50 milliseconds. During the transient operation, the target sliding window parameter is set to a maximum truncation delay of 5 milliseconds in step three. Since the typical transmission delay of a cellular network link is 30 to 80 milliseconds, only packets on the fiber optic link can be collected within the 5-millisecond truncation window; all packets on the cellular link fall outside the window.

[0103] In the truncation determination method of the first embodiment, a timeout eviction is directly performed with a threshold of 5 milliseconds, and the truncated set only contains fiber optic link packets. Since the edge terminal performs a deep copy of the packet payload and then sends a complete copy through both channels, both the fiber optic link and the cellular link independently carry the full sampled data. Within the 5-millisecond truncation window, all cellular link packets fall outside the window due to transmission delays far exceeding the window threshold, but the fiber optic link packets are sufficient to independently provide complete physical temporal coverage. In this dual-transmit, selective-receive fully redundant mode, physical temporal coverage gaps only occur when packet loss or congestion also occurs in the fiber optic link itself.

[0104] It should be noted that although the edge intelligent terminal performs a deep copy of the message payload and uploads it concurrently through dual channels, under ideal conditions the fiber optic link can independently provide complete temporal coverage. However, in actual power distribution network operation environments, fiber optic Ethernet links are not absolutely lossless. Millisecond-level message loss or delays may occur in power fiber optic ring networks during scenarios such as maintenance loop reversal, self-healing switching of fiber optic cables after interruption, OTN cross-connection reconstruction, and error retransmission triggered by optical power attenuation. Furthermore, when the edge intelligent terminal operates in high-frequency sampling mode, the bandwidth utilization of the fiber optic Ethernet port can reach a high level. During burst peaks where multiple terminals simultaneously report transient data, buffer overflows at the ring network switch port can also cause some fiber optic messages to be discarded. Therefore, even in a dual-channel fully redundant transmission architecture, physical temporal coverage holes may still exist in the truncated set, making a physical temporal coverage density sensing mechanism necessary in engineering practice.

[0105] In the truncation determination method of the second embodiment, a preliminary screening and truncation is first performed with a threshold of 5 milliseconds to form a preliminary screening and truncation message set, while retaining the queue of residual candidate messages. Then, physical temporal coverage density analysis is performed on the preliminary screening set. The micro-interval time width is set to 0.5 milliseconds, and the 50-millisecond-long tripping process is divided into 100 micro-intervals. The minimum fidelity density threshold for each micro-interval is set to 3 messages, meaning that under a reference sampling frequency of 10kHz, there should be at least 5 sampling points within a 0.5-millisecond micro-interval; 3 is taken as the minimum threshold to retain a certain fault tolerance margin. When the coverage density of some micro-intervals is found to be below one, the system targets and retrieves cellular link messages for the corresponding time period from the queue of residual candidate messages for supplementary sampling and fusion. If the residual void after supplementary sampling is an empty set, the physical interpolation enable signal is downgraded to an invalid state, and the entire numerical integration stage of the differential equation is skipped, saving cloud floating-point computing resources.

[0106] In the physical interpolation stage, the electro-magnetic-mechanical coupled differential equations of the power distribution switch include coil circuit equations and mechanical motion equations. The coil circuit equations describe the relationship between control voltage, coil resistance, inductance as the core travels, and the induced electromotive force. The mechanical motion equations describe the relationship between electromagnetic force, spring reaction force, frictional resistance, and core acceleration according to Newton's second law. The fourth-order Runge-Kutta algorithm has an integration step size of 0.01 milliseconds, meaning it performs 10 intermediate integrations within a 0.1 millisecond sampling interval corresponding to a reference sampling frequency of 10 kHz, ensuring numerical stability. For a typical 5-millisecond missing period, the Runge-Kutta solver needs to perform approximately 500 integration steps, generating 50 virtual message frames to fill the missing segment.

[0107] In the reordering stage, for the mixed set containing real fiber optic link messages and Runge-Kutta predicted virtual messages, a merge sort algorithm is used to sort them in ascending order with the absolute timestamp as the primary key. After sorting, the output high-fidelity time-series reconstruction sequence has a time interval between each frame that is strictly equal to the baseline sampling period of 0.1 milliseconds, and there are no time domain discontinuities, which can be directly injected into the downstream digital twin model.

[0108] Step five involves injecting the reconstructed high-fidelity time-series sequence into the digital twin model for virtual-real mapping and hybrid diagnostics, outputting a health diagnosis result for the distribution switch that includes a health score and a predicted remaining lifespan. It should be understood that the reconstructed high-fidelity time-series sequence itself is merely raw observation data that has undergone spatiotemporal alignment; it has not yet extracted information on the degradation status of the distribution switch control equipment or fault precursor characteristics. To achieve the ultimate goal of remote operation and maintenance management, and to quantitatively assess the health status of the distribution switch and predict its remaining lifespan, this high-fidelity sequence needs to be injected into a cloud-based digital twin model for in-depth analysis.

[0109] Digital twin models comprehensively consider the mechanical characteristic curves of power distribution switches, the aging patterns of electrical insulation, and a historical fault mode database, enabling dynamic estimation of the degree to which the current state of equipment deviates from the normal baseline. By comparing actual observation data with the expected output of the virtual model, the degradation deviation of the equipment can be quantified. Furthermore, neither a single physical mechanism model nor a single data-driven model can independently achieve accurate health diagnosis: while physical mechanism models have clear causal relationships, they are limited by dimensionality reduction simplification assumptions; while data-driven models can capture latent degradation features, they are prone to misjudgment under small sample conditions. Therefore, a hybrid diagnostic architecture integrating physical mechanism models and deep learning networks is needed. A Bayesian confidence fusion framework is used to comprehensively determine the diagnostic conclusions of both, outputting reliable health scores and remaining life predictions.

[0110] In one embodiment of this application, step five includes: performing virtual-real mapping and state forward extrapolation on the time-series reconstructed high-fidelity sequence to obtain the switch physical residual map and twin forward state feature vector; and performing fault posterior correction and health weighted scoring on the time-series reconstructed high-fidelity sequence and switch physical residual map to obtain the distribution switch health diagnosis result.

[0111] First, the high-fidelity time-series reconstruction sequence is injected into a cloud-based digital twin engine. The twin engine performs a virtual-to-real state mapping comparison based on the historical degradation baseline of the power distribution switch, calculates the Euclidean norm 2 deviation between the expected output of the virtual model under ideal baseline parameters and the current actual reconstruction sequence, and generates a physical residual feature map to characterize the degree of physical degradation of the equipment.

[0112] The calculation method for this residual extraction is as follows:

[0113] in, To calculate the generated physical residual feature map, The ideal physical mechanism baseline mapping function is built into the digital twin engine. The calibration history degradation baseline parameters are used to determine the equipment's condition at the factory or in a brand new, healthy state. High-fidelity time series reconstructed from actual observation input. The Euclidean L2 norm is used to quantify the absolute deviation distance.

[0114] Subsequently, using the nonlinear state transition matrix built into the digital twin engine, combined with the current physical state observations and system control inputs, forward integration is performed in phase space to solve for the next time slice. The resulting future predicted state is then tensor-concatenated with the current physical residual to construct the twin forward state feature vector.

[0115] The calculation method for this forward inference and cascading is as follows:

[0116] in, This is the output twin forward state feature vector. To characterize the nonlinear state transition operator of the system dynamics, This refers to the external control input applied to the power distribution switch at the current moment. For tensors, the concatenation operator is used on the feature dimension. This is a dimensionality reduction tiling operation that flattens a multidimensional feature map into a one-dimensional vector.

[0117] Secondly, the high-fidelity time-series reconstructed sequence is input into a deep learning fault classification network, and after feature mapping, a fault prior probability distribution from a purely data-driven perspective is output. In one embodiment of this application, fault posterior correction and health weighted scoring are performed on the high-fidelity time-series reconstructed sequence and the switch physical residual map to obtain the distribution switch health diagnosis result. This includes: using a deep learning fault classification network to perform implicit degradation feature mapping on the high-fidelity time-series reconstructed sequence to obtain a data-driven fault prior probability distribution; based on a Bayesian confidence fusion framework, using the switch physical residual map as a likelihood condition to perform posterior probability correction on the data-driven fault prior probability distribution to obtain a fused fault posterior probability distribution; performing weighted summation and inverse health mapping on the fused fault posterior probability distribution according to the severity penalty weights of each fault mode to obtain the distribution switch health diagnosis result; and generating a distribution switch sampling control command based on the health score in the distribution switch health diagnosis result through negative exponential decay mapping and feeding it back to step one. Specifically, deep learning networks are responsible for mining latent degradation features from massive historical operational data, enabling them to identify complex degradation patterns not modeled by the physical mechanism model. A Bayesian confidence fusion framework is introduced, using data-driven prior probabilities and physical residual feature maps as likelihood conditions for posterior probability fusion calculations. The physical mechanism model is responsible for establishing the causal chain of equipment degradation processes; its output residual map reflects the degree to which the actual state of the equipment deviates from the expected physical laws, effectively correcting the overconfidence that deep learning networks may exhibit when facing small samples or out-of-distribution operating conditions.

[0118] The calculation method for Bayesian fusion is as follows:

[0119] in, For the equipment to experience a type i fault under the combined conditions of the observation sequence and physical residuals. The posterior probability, Refers to the high-fidelity temporal reconstruction sequence of the input. Refers to the physical residual characteristic spectrum. For a deep learning network, the prior confidence probability of the i-th type of fault is given only based on the data sequence. Let C be the conditional likelihood probability of the current physical residual map given that the i-th type of fault occurs, and let C be the total number of potential fault mode categories defined in the system knowledge base.

[0120] Based on the fused multi-class fault posterior probabilities, a weighted summation is performed using the severity penalty weights of each fault mode, followed by inverse mapping, to calculate the current health score. The health score is calculated as follows:

[0121] in, Rate the device's health; a maximum score of 100 indicates the device is in brand new, healthy condition. is the severity penalty weight coefficient for the i-th type of failure mode.

[0122] Furthermore, based on the historical time series of health scores, an exponential degradation trend extrapolation model is used to predict the remaining service life of the power distribution switches. Specifically, the cloud-based operation and maintenance management platform maintains the health scores calculated for each power distribution switch over multiple consecutive operation and maintenance cycles, arranging them chronologically to form a health degradation trajectory. Assuming that the equipment health follows an exponential decay degradation law with cumulative operating time, the least squares method is used to fit the parameters of the historical degradation trajectory to obtain the exponential decay parameters characterizing the degradation rate of individual equipment. The fitted degradation curve is extrapolated towards future time to calculate the remaining operating time corresponding to the first drop in the health score to the system-defined critical scrapping limit, which is used as the predicted value of the remaining service life.

[0123] The remaining life prediction is calculated as follows:

[0124]

[0125] in, This is an exponential degradation model of health over cumulative running time t. This serves as the initial health score baseline for the equipment during its initial operation or after the most recent maintenance and recovery. RUL is the exponential degradation rate parameter obtained by fitting the historical health degradation trajectory using the least squares method, and RUL is the predicted remaining useful life. The health score is calculated based on the current maintenance cycle. The critical scrapping limit is defined by the system, which is the health threshold value at which the equipment no longer meets the requirements for safe operation. The health score and the remaining service life prediction value are encapsulated to generate the health diagnosis result of the power distribution switch.

[0126] Finally, the health score in the health diagnosis results of the power distribution switch is analyzed in depth. A mapping function with negative exponential decay characteristics is used to transform the health score into a degradation level warning weight coefficient for edge intelligent terminals. When the health score approaches the danger lower limit, the weight coefficient increases exponentially, driving the edge terminal to increase the sampling frequency to enhance its ability to capture rapid degradation features.

[0127] The mapping function is calculated as follows:

[0128] in, The generated degradation level warning weight coefficient is the core load value of the distribution switch sampling and control command. Rate the current health of the device. and These represent the upper limit of the safe maximum score and the lower limit of the critical scrapping threshold for the equipment health rating as defined by the system. This is the base gain scaling factor for the weighting instruction. A fast decay factor to control the degree of exponential decay.

[0129] The calculated weighting coefficients are packaged into instruction serialization according to the edge-side communication protocol specification to generate power distribution switch sampling and control instructions, which are then sent to the edge intelligent terminal and fed back to step one to form a cloud-edge closed-loop feedback. Simultaneously, the twin forward state feature vector is fed back to step three for use in the next cycle's operating condition identification.

[0130] In one specific embodiment, the degradation baseline parameters of the digital twin model are derived from the standard mechanical characteristic test recording data of the distribution switch during the factory acceptance phase, including standard coil current curves, standard contact travel-time curves, and standard spring force-displacement curves. During the virtual-real mapping simulation, the twin engine compares the high-fidelity sequence collected in the current maintenance cycle with the aforementioned baseline curves point by point. The output physical residual feature map intuitively reflects the degree of deviation of the equipment under degradation modes such as increased coil resistance, spring fatigue relaxation, and increased mechanical friction.

[0131] The deep learning fault classification network adopts a one-dimensional convolutional residual network architecture, containing five convolutional layers and two fully connected layers. The input length is 500 sampling points, corresponding to a physical action process of 50 milliseconds. The network obtains parameters through offline training on tens of thousands of historical closing and opening operation waveform data, and can identify a variety of fault modes, including coil inter-turn short circuits, core jamming, spring breakage precursors, contact erosion, and insulation partial discharge. The total number of potential fault mode categories C defined in the system knowledge base is set to 8, covering common mechanical, electrical, and insulation faults in distribution switches.

[0132] The severity penalty weighting coefficients for each fault mode are set differently based on the severity of the fault consequences. For faults such as core jamming that may lead to failure to operate or malfunction, the weighting coefficient is set to 95; for faults such as spring fatigue, the weighting coefficient is set to 70; and for faults such as slight contact wear, the weighting coefficient is set to 30. The general design principle for each weighting coefficient is: the more severe the fault consequences, the greater the penalty on the health score.

[0133] In the closed-loop feedback loop, the maximum safe score for health rating is set to 100, and the minimum critical failure score is set to 20. The base gain scaling factor is set to 1.0, and the agility decay factor is set to 3.0. Under this parameter configuration, when the device health rating is 90 (good condition), the weighting coefficient is approximately 0.05, which is lower than the status degradation alarm threshold of 0.6 in step one, and the edge terminal maintains the baseline sampling frequency. When the device health rating drops to 50 (significant degradation), the weighting coefficient is approximately 0.39, still lower than the alarm threshold. When the health rating further drops to 35 (severe degradation), the weighting coefficient is approximately 0.64, exceeding the alarm threshold of 0.6, and the edge terminal automatically switches to a high-frequency sampling mode for encrypted monitoring. This exponential decay characteristic ensures that the system only increases the sampling frequency when the device truly enters the accelerated degradation phase, avoiding resource waste on healthy devices.

[0134] For remaining lifetime prediction, the system maintains a historical sequence of health scores for each distribution switch for the most recent 60 maintenance cycles (approximately 60 days, based on a complete diagnostic calculation performed daily). The least squares method is used to fit an exponential function to this sequence to obtain the degradation rate parameter. The critical scrapping threshold is set at 20 points, consistent with the lower limit in the closed-loop feedback. For example, if a distribution switch currently has a health score of 65 points and the fitted degradation rate parameter is 0.003 / day, then the predicted remaining service life is... This predicted value is dynamically adjusted based on the health score update for each operational cycle, and is updated when degradation accelerates. Increased lifespan leads to a shorter predicted lifespan, even when the degradation trend slows. The reduction leads to an extended predicted lifespan. When equipment undergoes maintenance and its health score shows a step improvement, the system uses the initial score after the maintenance as the new score. The baseline value restarts the degenerate trajectory fitting process.

[0135] In summary, the remote operation and maintenance management method for power distribution switch control equipment based on the embodiments of this application is explained. It continuously tracks the real-time latency deviation between the cellular network and the fiber optic Ethernet dual-channel in the cloud as a reference parameter on the network side. Simultaneously, it utilizes the forward state extrapolation capability of the digital twin model to perceive in real time whether the power distribution switch is in a stable, inactive state or a transient state of closing / opening operation. These two factors are then combined as the basis for deciding on the expansion and contraction of the sliding window. When the equipment is in a stable monitoring period, the window is allowed to fully expand to maximize the inclusion of network disorder and ensure data integrity. When the equipment enters a transient sensitive period of millisecond-level rapid action, the window is forcibly contracted to ensure ultimate real-time response. For message periods inevitably missing due to window contraction, the differential equations of the electro-magnetic-mechanical coupling physical mechanism of the power distribution switch are invoked for phase space forward interpolation to bridge time-domain discontinuities. Finally, the high-fidelity time-series data, after disorder repair and physical interpolation, is injected into the digital twin model to complete virtual-real mapping and hybrid health diagnosis.

[0136] Figure 6 This is a schematic block diagram of a remote operation and maintenance management system for power distribution switch control equipment according to an embodiment of this application, such as... Figure 6As shown, the remote operation and maintenance management system 600 for power distribution switch control equipment includes: an edge acquisition and encapsulation module 610, used to synchronize and encapsulate multi-source operating data of the power distribution switch according to the frequency assigned by the sampling and control command of the power distribution switch using an edge intelligent terminal, and upload it concurrently through dual channels of cellular network and fiber optic Ethernet to obtain a dual-link message stream; a delay analysis and caching module 620, used to estimate the dual-channel transmission delay deviation online based on the difference between the absolute timestamp and the arrival time in the dual-link message stream to obtain the link delay difference reference value, and cache it as an out-of-order message queue according to the arrival order; and a twin state perception module 630, used to detect the closing and opening and current change indicators in the out-of-order message queue and the twin forward state feature vector. The system identifies the operating conditions of the power distribution switch. In the steady state without operation, the sliding window is enlarged using the link delay difference reference value. In the transient state with operation, the sliding window is forcibly shrunk using a preset limit truncation delay to obtain the target sliding window parameters and the physical interpolation enable signal. The timing reconstruction and interpolation module 640 is used to truncate and dequeue out-of-order messages according to the target sliding window parameters and rearrange them in ascending order according to the timestamp. When the physical interpolation enable signal is valid, it performs physical phase space dynamic interpolation on messages with missing time periods to obtain a high-fidelity timing reconstruction sequence. The hybrid diagnostic module 650 is used to inject the high-fidelity timing reconstruction sequence into the digital twin model for virtual-real mapping deduction and hybrid diagnostics to output the power distribution switch health diagnostic results, which include health scores and remaining life prediction values.

[0137] Here, those skilled in the art will understand that the specific operations of each step in the above-described method for remote operation and maintenance management of power distribution switch control equipment have been referenced above. Figures 1 to 5 The description of the remote operation and maintenance management system for power distribution switch control equipment is detailed here, and therefore, its repeated description will be omitted.

[0138] The terms "first," "second," etc., are used only to distinguish different technical features and do not imply their importance or number. The designation of a feature as "first" or "second" does not preclude the possibility of additional similar features. The scope of protection of this application shall be determined by the claims; any modifications, substitutions, or combinations that do not depart from the core idea of ​​this application shall fall within the scope of protection of this application.

[0139] The above description is merely a preferred embodiment of this application and is not intended to limit this application in any way. Although this application has been disclosed above with reference to preferred embodiments, it is not intended to limit this application. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the technical solution of this application. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for remote operation and maintenance management of power distribution switch control equipment, characterized in that, include: Step 1: Using the edge intelligent terminal, the multi-source operation data of the power distribution switch is synchronized and message encapsulated according to the frequency assigned by the sampling and control command of the power distribution switch. The data is then uploaded concurrently through the dual channels of cellular network and fiber optic Ethernet to obtain a dual-link message stream. Step 2: Based on the difference between the absolute timestamp and the arrival time in the dual-link message stream, estimate the dual-channel transmission delay deviation online to obtain the link delay difference benchmark value, and cache it as an out-of-order message queue according to the arrival order. Step 3: Based on the closing and opening circuit breaker and current change identifiers and twin forward state feature vectors in the out-of-order message queue, identify the operating conditions of the distribution switch. In the steady state without operation, the sliding window is enlarged by the link delay difference reference value. In the transient state with operation, the sliding window is forcibly shrunk by the preset limit truncation delay to obtain the target sliding window parameters and physical interpolation enable signal. Step 4: Truncate and dequeue the out-of-order message queue according to the target sliding window parameters and rearrange them in ascending order of timestamp. When the physical interpolation enable signal is valid, perform physical phase space dynamic interpolation on the messages in the missing time period to obtain a high-fidelity sequence for time-series reconstruction. Step 5: Inject the high-fidelity time-series reconstruction sequence into the digital twin model for virtual-real mapping and hybrid diagnosis to output the health diagnosis results of the power distribution switch, which include health scores and remaining life prediction values.

2. The remote operation and maintenance management method for power distribution switch control equipment according to claim 1, characterized in that, The multi-source operating data of the power distribution switch includes the current waveform of the closing and opening coils, the displacement of the energy storage spring, the insulation resistance value, and the ambient temperature and humidity values.

3. The remote operation and maintenance management method for power distribution switch control equipment according to claim 2, characterized in that, Step one includes: Based on the analog-to-digital conversion frequency of the dynamically modulated deterioration warning weight coefficient parsed from the sampling and control command of the power distribution switch, the multi-source operation data of the power distribution switch is synchronously sampled at high frequency and spliced ​​with discrete sequences to obtain the digital feature tuple of the switch. Using the network path delay offset and local crystal oscillator drift error measured by the precise clock synchronization protocol, the switch digital feature tuple is frame-by-frame absolute time-stamp calibration and frame header writing are performed to obtain the switch timing feature tuple. After encapsulating the switch timing feature tuple with a unique device identifier, it is uploaded concurrently via cellular network and fiber optic Ethernet to obtain a dual-link message stream.

4. The remote operation and maintenance management method for power distribution switch control equipment according to claim 1, characterized in that, Step two includes: Physical payload extraction and arrival time attribute projection are performed on dual-link message streams to obtain dual-timestamped message sequences; Online link delay jitter estimation is performed on double-time-stamped message sequences to obtain a baseline value for the link delay difference and a set of time-stamped messages; By pushing packets into a first-in-first-out circular buffer in ascending order of arrival time, the set of time-stamped packets is physically cached sequentially and logically instantiated out of order to obtain an out-of-order packet queue.

5. The remote operation and maintenance management method for power distribution switch control equipment according to claim 1, characterized in that, Step three includes: Lightweight window scanning and mutation feature extraction are performed on the out-of-order message queue to obtain the switch local mutation feature vector; The virtual and real states are jointly spliced ​​and transient probability inferred from the switch local mutation feature vector and twin forward state feature vector to obtain the transient probability score of the operating condition. Based on the decision result between the transient probability score of the operating condition and the preset service sensitivity threshold, the window scaling and interpolation enable settings are applied to the transient probability score of the operating condition and the reference value of the link delay difference to obtain the target sliding window parameters and the physical interpolation enable signal.

6. The remote operation and maintenance management method for power distribution switch control equipment according to claim 1, characterized in that, Step four includes: Based on the comparison of the difference between the current system time and the message arrival time with the target sliding window parameters, the out-of-order message queue is subjected to waiting timeout determination and forced dequeue truncation to obtain the truncated message set with timeout. When the physical interpolation enable signal is valid, the missing time period state forward interpolation and virtual message union fusion are performed on the truncated time-limited message set to obtain the interpolated fused extended message set. Using the absolute timestamp on the edge side as the unique sorting key, the interpolated fusion extended message set is subjected to causal temporal rearrangement and linear continuous solidification to obtain a high-fidelity temporal reconstruction sequence.

7. The remote operation and maintenance management method for power distribution switch control equipment according to claim 1, characterized in that, Step five includes: Virtual-real mapping and state forward inference are performed on the high-fidelity time-series reconstruction sequence to obtain the switch physical residual map and twin forward state feature vector; Fault posterior correction and health weighted scoring are performed on the high-fidelity sequence of time-series reconstruction and the physical residual map of the switch to obtain the health diagnosis results of the distribution switch.

8. The remote operation and maintenance management method for power distribution switch control equipment according to claim 6, characterized in that, Based on a step-by-step comparison of the difference between the current system time and the message arrival time with the target sliding window parameters, the out-of-order message queue is subjected to timeout determination and forced dequeueing to obtain a truncated message set with timeout, including: Based on the comparison of the difference between the current system time and the arrival time of each message with the target sliding window parameters, the out-of-order message queue is divided into timeout eviction and in-transit retention to obtain the initial truncated message set and the queue residual candidate message set. The temporal coverage density calculation and low-density hole identification are performed on the initial screening and truncated message set to obtain the temporal coverage hole interval set; Based on the set of time-domain coverage gaps, the set of residual candidate messages in the queue is fused with timestamp-targeted retrieval and early dequeueing to obtain the set of truncated time-limited messages.

9. The remote operation and maintenance management method for power distribution switch control equipment according to claim 7, characterized in that, Fault posterior correction and health-weighted scoring are performed on the high-fidelity time-series reconstruction sequence and the physical residual map of the switches to obtain the health diagnosis results of the distribution switches, including: A deep learning fault classification network is used to perform implicit degradation feature mapping on high-fidelity time-series reconstruction sequences to obtain a data-driven fault prior probability distribution. Based on the Bayesian confidence fusion framework, the posterior probability correction of the data-driven fault prior probability distribution is performed using the switch physical residual map as a likelihood condition to obtain the fused fault posterior probability distribution. The weighted summation of the fused fault posterior probability distribution is performed based on the severity penalty weight of each fault mode, and then inversely mapped to the health status to obtain the health diagnosis result of the distribution switch. Based on the health score in the health diagnosis result of the distribution switch, a sampling control command for the distribution switch is generated and fed back to step one through negative exponential decay mapping.

10. A remote operation and maintenance management system for power distribution switch control equipment, characterized in that, include: The edge acquisition and encapsulation module is used to synchronize and encapsulate the multi-source operation data of the power distribution switch according to the frequency assigned by the sampling and control command of the power distribution switch using the edge intelligent terminal, and then upload it concurrently through the dual channels of cellular network and fiber optic Ethernet to obtain a dual-link message stream. The delay analysis and caching module is used to estimate the dual-channel transmission delay deviation online based on the difference between the absolute timestamp and the arrival time in the dual-link message stream to obtain the link delay difference benchmark value, and cache it as an out-of-order message queue according to the arrival order. The twin state perception module is used to identify the operating conditions of the power distribution switch based on the closing and opening and current change identifiers in the out-of-order message queue and the twin forward state feature vector. In the steady state without action, the sliding window is enlarged by the link time delay difference reference value, and in the transient state with action, the sliding window is forcibly shrunken by the preset limit truncation time delay to obtain the target sliding window parameters and physical interpolation enable signal. The timing reconstruction and interpolation module is used to truncate and dequeue out of the out-of-order message queue according to the target sliding window parameters and rearrange them in ascending order of timestamps. When the physical interpolation enable signal is valid, it performs physical phase space dynamic interpolation on the messages in the missing time period to obtain a high-fidelity timing reconstruction sequence. The hybrid diagnostic module is used to inject high-fidelity time-series reconstruction sequences into a digital twin model for virtual-real mapping and hybrid diagnostics, so as to output health diagnostic results for distribution switches that include health scores and remaining life prediction values.

Images