Method and system for guaranteeing full-period operation and maintenance of direct current system

Abstract

The invention discloses a full-period operation and maintenance guarantee method and system for a direct current system, and the method comprises the steps: collecting system state data in real time through a multi-source sensor network, and carrying out the synchronization and normalization processing, and then generating a standardized time series data set; utilizing the pre-trained fault prediction neural network model to carry out probabilistic evaluation on future fault risks; when the risk exceeds a threshold value, an isolation decision engine is triggered to generate an optimal isolation instruction according to system topology and a real-time state, and a solid-state switch and a mechanical mechanism in the intelligent execution unit cooperate to complete millisecond-level rapid isolation; after isolation, a closed-loop feedback module collects steady-state parameters and compares the steady-state parameters with expected values, and online self-adaptively optimizes a prediction model and decision parameters; during normal operation of the system, the preventive maintenance module automatically generates a digital list based on the equipment health score. According to the method, full-period intelligent operation and maintenance of the direct current system from passive response to active prediction, automatic disposal and continuous optimization are realized, and the reliability and the safety of the system are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and in particular to a method and system for ensuring the full life cycle operation and maintenance of a DC system. Background Technology

[0002] DC systems are critical infrastructure in power systems, rail transit, data centers, and new energy power generation, and their operational stability is paramount. Currently, the operation and maintenance of DC systems mainly rely on manual inspections and reactive responses. While existing technologies have introduced intelligent inspection terminals for data collection, their functions are limited to status monitoring and alarms, lacking the ability to predict faults and an automated closed-loop handling mechanism. This results in delayed operation and maintenance responses, easy spread of faults, and system availability that fails to meet the demands of modern high-reliability applications.

[0003] Therefore, there is an urgent need in this field for a full-cycle operation and maintenance guarantee solution for DC systems that can achieve proactive fault prediction, automatic isolation, and self-optimization capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for full-cycle operation and maintenance of DC systems, which aims to solve the technical problems of passive operation and maintenance, delayed fault response, and lack of closed-loop optimization in existing DC systems, realize the transformation from post-event maintenance to pre-event prevention, and from manual handling to automatic closed-loop, and improve the reliability and operation and maintenance efficiency of DC systems.

[0005] To address the aforementioned problems, according to one aspect of this application, an embodiment of the present invention provides a method for ensuring the full lifecycle operation and maintenance of a DC system, comprising:

[0006] S1: Through a multi-source sensor network deployed at key nodes of the DC system, real-time electrical, thermodynamic and communication status data of the system are collected to form a multi-dimensional real-time operating status data stream covering the entire topology of the system;

[0007] S2: Perform time-series synchronization and normalization processing on the multi-dimensional real-time running status data stream to generate a standardized time-series dataset containing multiple feature sequences;

[0008] S3: Based on the standardized time series dataset, call the pre-trained fault prediction model to perform a probabilistic assessment of the potential fault risks of the DC system within a future preset time window, and output the fault occurrence probability value of each key device.

[0009] S4: When the probability of failure of any critical equipment exceeds the preset threshold, the fault isolation decision engine is triggered to generate the optimal fault isolation path instruction based on the system topology, load distribution, backup capacity and equipment health status.

[0010] S5: Send the optimal fault isolation path instruction to the corresponding intelligent execution unit to control it to complete the electrical disconnection and physical isolation operation of the fault branch;

[0011] S6: After completing the fault isolation operation, collect the steady-state operating parameters of the isolated system and calculate the deviation with the expected performance indicators. Based on the calculation results, perform online adaptive optimization of the parameters of the fault prediction model and the fault isolation decision engine.

[0012] S7: During normal system operation, preventive maintenance planning is periodically initiated, generating equipment-level maintenance priority ranking and digital work orders based on multi-dimensional equipment status information.

[0013] In some implementations, in step S2, the timing synchronization adopts an interpolation alignment method based on a unified time reference, and the normalization process adopts a maximum-minimum scaling method to compress each physical quantity data to a predetermined range; for data points that exceed the rated working range, they are marked as abnormal data and stored separately.

[0014] In some implementations, in step S3, the fault prediction model is a neural network model built on a long short-term memory network. Its input is a standardized time series dataset of historical time windows, and its output is the probability value of fault occurrence at different time granularities in the future. The model is trained through supervised learning using a sample library containing time series data of historical fault events.

[0015] In some implementations, in step S4, the fault isolation decision engine generates the optimal fault isolation path instruction by performing the following steps:

[0016] a. Identify the electrical island where the faulty device is located;

[0017] b. Calculate the set of all possible isolation boundary nodes within the electrical island;

[0018] c. Assess the impact of disconnecting each boundary node on the continuity of power supply to non-faulty areas and the complexity of the operation;

[0019] d. Select the combination of boundary nodes with the least impact and lowest operational complexity as the isolation execution point, and generate an instruction containing the sequence of switches to be operated and a list of load transfer priorities.

[0020] In some implementations, in step S5, the intelligent execution unit includes a programmable logic controller, a solid-state switch array, and a mechanical disconnect switch drive mechanism; after parsing the instruction, the programmable logic controller first controls the solid-state switch array to perform millisecond-level electrical disconnection according to a preset timing sequence, and then drives the mechanical disconnect switch drive mechanism to complete physical isolation.

[0021] In some implementations, in step S6, the deviation calculation adopts the weighted Euclidean distance formula to comprehensively evaluate the bus voltage recovery accuracy, load power supply continuity, standby path temperature rise rate and system total loss increment; when the deviation exceeds the tolerance threshold, online fine-tuning of the fault prediction model weight parameters and isolation decision engine path optimization algorithm parameters is triggered.

[0022] In some implementations, in step S7, the multi-dimensional status information of the equipment includes the cumulative running time of the equipment, historical fault records, environmental adaptability, maintenance response timeliness, and spare parts availability; the equipment-level maintenance priority ranking is determined by calculating the equipment health status score using a multi-index comprehensive evaluation method.

[0023] This invention also provides a DC system full-cycle operation and maintenance support system for implementing any of the above-described DC system full-cycle operation and maintenance support methods, including:

[0024] The multi-source status sensing module is used to collect real-time all-round operating status data of the DC system;

[0025] The data preprocessing module is used to perform time-series synchronization and normalization on the collected data to generate a standardized time-series dataset.

[0026] The fault prediction module is used to perform a probabilistic assessment of potential fault risks by calling a pre-trained fault prediction model based on the standardized time series dataset.

[0027] The isolation decision execution module is used to generate and issue the optimal fault isolation path instruction when the fault risk exceeds the threshold, and control the intelligent execution unit to complete the isolation operation.

[0028] A closed-loop feedback optimization module is used to adaptively optimize the parameters of the fault prediction module and the isolation decision execution module online based on the performance deviation of the isolated system.

[0029] The preventive maintenance planning module is used to periodically generate equipment maintenance priority rankings and digital work orders during normal system operation.

[0030] In some embodiments, the intelligent execution unit includes a programmable logic controller, a solid-state switch array based on a wide bandgap semiconductor device, and a mechanical disconnect switch drive mechanism with high-precision position feedback. The programmable logic controller is used to coordinate and control the timing operations of the solid-state switch array and the mechanical disconnect switch drive mechanism.

[0031] In some implementations, a human-machine interface is also included to graphically display the system topology, equipment health status heatmap, fault risk prediction curve, isolation operation status and maintenance work order records in real time, and to support manual triggering of equipment self-test and temporary adjustment of isolation strategy.

[0032] Compared with the prior art, the DC system full-cycle operation and maintenance guarantee method and system of the present invention have at least the following beneficial effects:

[0033] Proactive: By using fault prediction models, risks can be identified before faults occur, thus shifting from "post-failure maintenance" to "pre-failure prevention".

[0034] Automation: Integrating decision-making and execution units enables automatic, rapid, and precise isolation of faults, significantly reducing manual intervention and response time.

[0035] Closed-loop and self-optimization: Through a closed-loop feedback mechanism, the system can continuously optimize its algorithm based on actual operating results, and has adaptive evolution capabilities.

[0036] Full lifecycle management: It organically combines daily monitoring, fault handling and preventive maintenance, covering the entire life cycle of equipment, and comprehensively improving system reliability and operation and maintenance efficiency.

[0037] High reliability: The isolation scheme, which combines solid-state switches and mechanical switches, balances speed and reliability, ensuring foolproof isolation operation.

[0038] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the overall system architecture provided in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the principle framework of a fault prediction neural network model.

[0042] Figure 3 A flowchart illustrating the logic of multi-source state perception and data preprocessing;

[0043] Figure 4 A collaborative control framework diagram for fault isolation decision-making and execution;

[0044] Figure 5 A diagram illustrating the parameter adjustment mechanism for closed-loop feedback optimization;

[0045] Figure 6 A logic diagram for scoring and work order generation in preventive maintenance planning;

[0046] Figure 7 This is a schematic diagram illustrating the topology and isolation boundary analysis of a DC system.

[0047] Figure 8 This is a diagram showing the relationship between the human-computer interaction interface and system data linkage. Detailed Implementation

[0048] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments, structures, features, and effects according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "an embodiment" or "an embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0049] In the description of this invention, it should be clearly stated that the terms "first," "second," etc., in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence; the terms "vertical," "lateral," "longitudinal," "front," "rear," "left," "right," "up," "down," "horizontal," etc., indicate orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, and are merely for the convenience of describing this invention, and do not mean that the device or element referred to must have a specific orientation or position, and therefore should not be construed as a limitation of this invention.

[0050] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0051] like Figures 1-8 As shown, this embodiment of the invention provides a method for ensuring the full lifecycle operation and maintenance of a DC system, including:

[0052] S1: Through a multi-source sensor network deployed at key nodes of the DC system, real-time electrical, thermodynamic and communication status data of the system are collected to form a multi-dimensional real-time operating status data stream covering the entire topology of the system;

[0053] S2: Perform time-series synchronization and normalization processing on the multi-dimensional real-time running status data stream to generate a standardized time-series dataset containing multiple feature sequences;

[0054] S3: Based on the standardized time series dataset, call the pre-trained fault prediction model to perform a probabilistic assessment of the potential fault risks of the DC system within a future preset time window, and output the fault occurrence probability value of each key device.

[0055] S4: When the probability of failure of any critical equipment exceeds the preset threshold, the fault isolation decision engine is triggered to generate the optimal fault isolation path instruction based on the system topology, load distribution, backup capacity and equipment health status.

[0056] S5: Send the optimal fault isolation path instruction to the corresponding intelligent execution unit to control it to complete the electrical disconnection and physical isolation operation of the fault branch;

[0057] S6: After completing the fault isolation operation, collect the steady-state operating parameters of the isolated system and calculate the deviation with the expected performance indicators. Based on the calculation results, perform online adaptive optimization of the parameters of the fault prediction model and the fault isolation decision engine.

[0058] S7: During normal system operation, preventive maintenance planning is periodically initiated, generating equipment-level maintenance priority ranking and digital work orders based on multi-dimensional equipment status information.

[0059] In this embodiment, a multi-source sensor network is first deployed at key nodes of the DC system to collect electrical, thermodynamic, and communication status data in real time, covering the entire system topology to form a multi-dimensional data stream. This includes voltage sensors at the DC bus, current sensors at branch outlets, temperature sensors on heat-sensitive components such as power device heat sinks, insulation monitoring devices based on the unbalanced bridge method, and intelligent terminals with built-in communication status monitoring modules. All sensors are connected to the central data acquisition unit via industrial Ethernet or a fiber optic ring network, ensuring that the data timestamp error is less than 10μs, providing an accurate data source for subsequent analysis. Figure 1 , Figure 3 ).

[0060] The acquired multi-dimensional data stream needs to undergo time-series synchronization and normalization processing. A cubic spline interpolation alignment method based on a unified time reference is adopted, and resampling is performed at a sampling period of 10ms. Then, the maximum and minimum value scaling method is used to compress each physical quantity to a predetermined range, such as voltage data corresponding to 90%-110% of the rated voltage and current data corresponding to 0-150% of the branch rated current. Data exceeding the rated range is marked as abnormal and stored separately, generating a standardized time-series dataset containing voltage fluctuation feature sequences, current harmonic distortion sequences, etc. Figure 3 ).

[0061] Based on this standardized time-series dataset, a pre-trained fault prediction model was invoked. This model is a three-layer long short-term memory network structure. The input layer receives a sequence of 1000 points formed by 10 seconds of historical data. The hidden layer extracts voltage sag correlation features, current surge coupling features, and temperature rise cumulative effect features through a gating mechanism. The output layer outputs the probability values ​​of fault occurrence for each key device at three time granularities: 10 seconds, 30 seconds, and 60 seconds. The model is trained under supervision using a historical fault sample database containing events such as voltage collapse and current overload over the past five years. The accuracy on the test set exceeds 97.7%. Figure 2 ).

[0062] When the failure probability of any device exceeds a preset threshold, the fault isolation decision engine is activated. It first traverses the system connection matrix to identify the electrical island where the faulty device is located using a breadth-first search. Then, it calculates the set of all possible isolation boundary nodes within the island and evaluates the impact on the continuity of power supply in the non-faulty area after each node is disconnected (affected load power / total load power) and the operational complexity (number of switching actions required). Finally, it selects the node combination with the least impact and the lowest complexity and generates the optimal instruction containing the sequence of switches to be operated and a load transfer priority list. Figure 4 , Figure 7 ).

[0063] After the optimal instruction is issued to the intelligent execution unit, the programmable logic controller within the unit parses the instruction and first controls the solid-state switch array based on wide bandgap semiconductors to perform millisecond-level electrical disconnection (silicon carbide device turn-off time < 20μs). Subsequently, it drives the mechanical disconnect switch drive mechanism with high-precision position feedback (position accuracy ±0.5°) to complete physical isolation. The entire process takes < 200ms, ensuring that the faulty branch quickly disconnects from the main system. Figure 4 ).

[0064] After isolation is completed, the system waits to enter a steady state (bus voltage fluctuation < 0.5% for 5 seconds). The closed-loop feedback module collects steady-state parameters such as bus voltage recovery accuracy and load power supply continuity. It calculates the deviation from the expected value using a weighted Euclidean distance formula (voltage recovery accuracy weight 0.4, load continuity weight 0.3, etc.). If the deviation exceeds the tolerance threshold, the fault prediction model weights are fine-tuned through backpropagation, and the path optimization parameters of the isolation decision engine are updated. Figure 5 ).

[0065] During normal system operation, preventative maintenance planning is initiated periodically every 24 hours. Based on five dimensions—accumulated equipment runtime, historical fault records, environmental adaptability, maintenance response timeliness, and spare parts availability—a health status score is calculated using a multi-indicator comprehensive evaluation method. Equipment maintenance priorities are generated according to the score, and digital work orders containing maintenance details, spare parts lists, and safety procedures are pushed to the operation and maintenance platform. Work orders are transmitted using national cryptographic algorithms. Figure 6 ).

[0066] This process enables full-cycle management from data acquisition to fault handling and maintenance planning, identifies fault risks in advance to avoid delayed response, automatically isolates to reduce manual intervention, optimizes closed-loop to improve system adaptability, and comprehensively enhances the reliability and safety of DC system operation.

[0067] In some implementations, in step S2, the timing synchronization adopts an interpolation alignment method based on a unified time reference, and the normalization process adopts a maximum-minimum scaling method to compress each physical quantity data to a predetermined range; for data points that exceed the rated working range, they are marked as abnormal data and stored separately.

[0068] In this embodiment, when processing the multi-dimensional real-time running status data stream, a unified time benchmark is first used as a reference, and the cubic spline interpolation alignment method is used to synchronize the data of each channel. This method can effectively fill in missing sampling points, ensure the continuity and smoothness of the data in the time dimension, and avoid deviations in subsequent analysis due to data asynchrony.

[0069] The synchronized data stream is resampled at a preset sampling period of 10ms, converting the original data of different frequencies into a time sequence with a fixed frequency, providing a unified time granularity for subsequent model input. Normalization is then performed using the maximum-minimum scaling method, setting specific rated ranges for different physical quantities: voltage data corresponds to 90%-110% of the equipment's rated voltage, current data corresponds to 0-150% of the branch's rated current, temperature data corresponds to -10℃-100℃, insulation resistance data corresponds to 1MΩ-500MΩ, and communication packet loss rate data corresponds to 0-1. All physical quantities are compressed into the 0-1 range using the formula (data - minimum value) / (maximum value - minimum value), eliminating the interference of dimensional differences on model calculations.

[0070] For data points that exceed the above-mentioned rated range discovered during the processing, the system will automatically mark them as abnormal data and will not participate in subsequent fault prediction model reasoning and decision calculation. However, they will be stored separately in the abnormal event database as auxiliary evidence for subsequent fault diagnosis, making it easier for maintenance personnel to trace the root cause of the anomaly.

[0071] This process not only ensures the standardization of data in both time and numerical dimensions, providing high-quality input data for fault prediction models, but also effectively filters out abnormal data, reduces the interference of invalid information on system decisions, and improves the accuracy of subsequent fault assessment and isolation decisions.

[0072] In some implementations, in step S3, the fault prediction model is a neural network model built on a long short-term memory network. Its input is a standardized time series dataset of historical time windows, and its output is the probability value of fault occurrence at different time granularities in the future. The model is trained through supervised learning using a sample library containing time series data of historical fault events.

[0073] In this embodiment, the fault prediction model used for fault risk assessment is a neural network model built based on a long short-term memory network. Its overall architecture includes an input layer, three hidden layers, and an output layer. Figure 2 The standardized time-series dataset received by the input layer needs to cover 10 seconds of historical running data, corresponding to 1000 sampling points (calculated at a 10ms sampling period). The data content includes multi-dimensional information such as voltage fluctuation characteristics, current harmonic distortion characteristics, and temperature rise gradient characteristics to ensure that the model can comprehensively capture the system's operating status.

[0074] The three hidden layers each perform different feature extraction functions: The first layer focuses on extracting voltage sag-related features through a gating mechanism, with a forget gate controlling the weight of historical voltage fluctuations on the current state, an input gate filtering effective voltage deviation information, and an output gate determining the feature transmission strength; The second layer focuses on current mutation coupling features, using cell states to record the cumulative effect of current harmonic components over time, and reflecting the dynamic correlation between current distortion and load changes through the hidden state output; The third layer integrates heterogeneous features such as temperature rise gradient, insulation degradation trend, and communication packet loss rate, coordinating the time dependencies of different types of data, and achieving deep integration of multi-dimensional information.

[0075] The output layer is a fully connected layer with three neurons corresponding to three time granularities: 10s, 30s, and 60s. It outputs the probability value of failure of each key device within the corresponding time window, providing the system with risk warnings with different timeframes.

[0076] The model training phase utilizes a sample library containing fault data from similar DC systems over the past five years for supervised learning. This library covers typical fault events such as voltage collapse, current overload, insulation breakdown, and communication interruption. Each sample is labeled with the fault type, occurrence time, and impact range. The training process employs a cross-entropy loss function to measure the difference between the predicted probability and the actual fault label. A stochastic gradient descent algorithm with a driving term is used to optimize the model parameters. The initial learning rate is set to 0.001. After each training round, the learning rate decay coefficient is dynamically adjusted based on the accuracy on the validation set (the learning rate is multiplied by 0.9 when the accuracy improvement is <0.5% for three consecutive rounds), ensuring the model converges to its optimal state. Ultimately, a fault prediction accuracy of 97.7% is achieved on the test set, with a false alarm rate controlled below 3.5%.

[0077] Leveraging the superior processing capabilities of long short-term memory networks for time-series data, this model can accurately capture potential fault trends in changes in system operating status. By providing early warnings to the system through multi-time-granularity probability outputs, it buys sufficient time for subsequent fault isolation and significantly reduces system losses caused by sudden faults.

[0078] In some implementations, in step S4, the fault isolation decision engine generates the optimal fault isolation path instruction by performing the following steps:

[0079] a. Identify the electrical island where the faulty device is located;

[0080] b. Calculate the set of all possible isolation boundary nodes within the electrical island;

[0081] c. Assess the impact of disconnecting each boundary node on the continuity of power supply to non-faulty areas and the complexity of the operation;

[0082] d. Select the combination of boundary nodes with the least impact and lowest operational complexity as the isolation execution point, and generate an instruction containing the sequence of switches to be operated and a list of load transfer priorities.

[0083] In this embodiment, when the probability value of a critical equipment failure output by the fault prediction model exceeds a preset threshold, the fault isolation decision engine immediately initiates its workflow. Figure 4 , Figure 7 First, the engine calls the DC system topology diagram stored in the system (stores the electrical connection relationships between devices in the form of an adjacency matrix, where matrix element 1 represents a direct connection and 0 represents no direct connection). Starting from the faulty device node, it traverses the entire topology network using a breadth-first search algorithm to identify all sets of devices that have an electrical connection relationship with the faulty device. This set is the electrical island where the faulty device is located, thus defining the initial scope of the fault's impact.

[0084] Subsequently, based on the electrical island topology, the engine calculates the set of all possible isolation boundary nodes within the island. Boundary nodes are defined as devices or switches that simultaneously connect the faulty electrical island and the non-faulty area; these nodes are key control points for preventing fault propagation. For each candidate boundary node, the engine evaluates it from two dimensions: first, the degree of power supply impact, which measures the interference with the continuity of power supply to the non-faulty area by calculating the ratio of the total power of the affected load after the node is disconnected to the total load power of the system; second, the operational complexity, which is judged by the number of switches required to perform the isolation operation of the node; the fewer the number, the lower the complexity.

[0085] After evaluating all boundary nodes, the engine integrates the results of the two evaluations and selects the combination of boundary nodes with the least power supply impact and the lowest operational complexity as the final isolation execution point. Simultaneously, by combining the current system load distribution (the percentage of actual current in each branch relative to the rated current), standby path capacity margin (the difference between the maximum allowable current in the standby path and the currently allocated current), and equipment health status scores, it generates an optimal fault isolation path instruction containing a sequence of circuit breaker numbers to be disconnected, a sequence of bypass switch numbers to be closed, and a load transfer priority list (primary load priority, secondary priority, and tertiary allowable short-term interruption). This ensures that the instruction blocks the fault while minimizing the impact on normal system operation.

[0086] Through this decision-making process, the engine can quickly formulate a scientific and reasonable isolation plan after a fault warning, avoiding blind isolation that could lead to power outages in non-faulty areas, ensuring continuous power supply to the core load of the system, and improving the accuracy and efficiency of fault handling.

[0087] In some implementations, in step S5, the intelligent execution unit includes a programmable logic controller, a solid-state switch array, and a mechanical disconnect switch drive mechanism; after parsing the instruction, the programmable logic controller first controls the solid-state switch array to perform millisecond-level electrical disconnection according to a preset timing sequence, and then drives the mechanical disconnect switch drive mechanism to complete physical isolation.

[0088] In this embodiment, after receiving the optimal fault isolation path instruction, the programmable logic controller in the intelligent execution unit first parses the instruction, verifies the legality of the digital signature attached to the instruction, and after confirming that there are no errors, extracts the switching action sequence, timing requirements and load transfer rules, and converts the abstract instruction into an executable hardware control signal.

[0089] The programmable logic controller first sends control signals to the solid-state switch array according to a preset timing sequence. The solid-state switch array is constructed using silicon carbide wide bandgap semiconductor devices. Its on-resistance is less than 5mΩ, and it can withstand the instantaneous impact of 10 times the rated current. Its turn-off time is less than 20μs, which can complete the electrical disconnection of the fault branch in milliseconds, quickly block the propagation of fault current, and prevent the fault from escalating.

[0090] Within 50ms of the solid-state switch array completing the electrical disconnection, the programmable logic controller (PLC) sends an action command to the mechanical disconnector drive mechanism. This drive mechanism employs a permanent magnet synchronous motor with a worm gear reducer, and incorporates a Hall effect position sensor for high-precision position feedback (accuracy ±0.5°). Its drive torque reserve coefficient is no less than 1.5, ensuring stable operation even under extreme conditions such as low temperature and high humidity. Upon receiving the command, the drive mechanism moves the moving contact of the disconnector until it reaches the fully disconnected physical isolation position. At this point, the Hall effect position sensor feeds back a position confirmation signal to the PLC.

[0091] After receiving the solid-state switch cut-off confirmation signal and the mechanical isolating switch position confirmation signal, the programmable logic controller determines that the isolation operation is complete, generates an operation report including the operation start time, the completion time of each switch action, the final isolation status and abnormal alarm information, and uploads it to the central monitoring platform through an independent communication channel for maintenance personnel to view in real time.

[0092] The entire isolation process, from the issuance of the command to the completion of physical isolation, takes less than 200ms. Through the coordinated action of solid-state switches and mechanical switches, it achieves both rapid electrical disconnection of the faulty branch and ensures safety through physical isolation, preventing fault recurrence and significantly improving the speed and reliability of fault isolation.

[0093] In some implementations, in step S6, the deviation calculation adopts the weighted Euclidean distance formula to comprehensively evaluate the bus voltage recovery accuracy, load power supply continuity, standby path temperature rise rate and system total loss increment; when the deviation exceeds the tolerance threshold, online fine-tuning of the fault prediction model weight parameters and isolation decision engine path optimization algorithm parameters is triggered.

[0094] In this embodiment, after the fault isolation operation is completed, the closed-loop feedback module starts working immediately and first enters the system steady-state determination stage: continuously monitor the bus voltage fluctuation. When the bus voltage fluctuation amplitude is less than 0.5% of the equipment rated voltage and this state lasts for more than 5 seconds, the system is determined to enter a new steady-state operation stage. At this time, the collected parameters can truly reflect the system performance after the isolation operation.

[0095] After steady state is established, the module collects four key performance parameters: bus voltage recovery accuracy (absolute difference between actual bus voltage and target voltage / target voltage), load power supply continuity index (total load power continuously supplied after isolation / total load power before isolation), standby path temperature rise rate (average temperature rise rate of key nodes in the standby path within 10 seconds), and total system loss increment (the difference between total system loss after isolation and total loss before isolation).

[0096] Subsequently, the weighted Euclidean distance formula was used to calculate the deviation between these steady-state parameters and the expected performance indicators output by the fault prediction model before isolation. The weights of each parameter in the formula were set according to their impact on system operation: bus voltage recovery accuracy weight 0.4, load power supply continuity index weight 0.3, standby path temperature rise rate weight 0.2, and system total loss increment weight 0.1. The weighted calculation ensured that the key performance indicators played a dominant role in the deviation results.

[0097] If the calculated weighted Euclidean distance is greater than the preset tolerance threshold (0.15), the parameter optimization process is triggered: the deviation value is used as an error signal and transmitted layer by layer to each neuron of the fault prediction model through the backpropagation algorithm. The gradient of each weight parameter is calculated using the chain rule (the gradient calculation step size is set to 0.0001). The weight parameters are fine-tuned according to the gradient direction and magnitude, and the fine-tuning amplitude is controlled within 0.5% of the original value to avoid parameter mutations that could lead to model instability. At the same time, the path optimization algorithm parameters in the fault isolation decision engine are updated, and the load weight coefficient and operation complexity penalty coefficient of the power supply impact degree calculation formula are adjusted to make subsequent decisions more in line with the actual operating characteristics of the current system.

[0098] After parameter adjustments are completed, the module generates an optimization log, which records in detail the parameter values ​​before and after the adjustment, the basis for the adjustment, and the expected improvement effect, providing a traceable basis for subsequent system maintenance and optimization. Through this closed-loop feedback optimization process, the system can continuously improve itself based on actual operating results, enhance the accuracy of fault prediction and the rationality of isolation decisions, and strengthen long-term operational adaptability.

[0099] In some implementations, in step S7, the multi-dimensional status information of the equipment includes the cumulative running time of the equipment, historical fault records, environmental adaptability, maintenance response timeliness, and spare parts availability; the equipment-level maintenance priority ranking is determined by calculating the equipment health status score using a multi-index comprehensive evaluation method.

[0100] In this embodiment, during normal operation of the DC system, the preventive maintenance module automatically starts on a fixed 24-hour cycle. The start time is selected at 2:00 AM when the system load is at its lowest, so as to avoid the maintenance planning process occupying system operating resources and affecting the power supply of core loads.

[0101] After the module starts, it first reads the cumulative runtime of each device from the equipment ledger database and calculates the runtime stability score using the formula "Runtime Stability Score = 100 - (Cumulative Runtime / 100000) × 20". The longer the cumulative runtime, the lower the score, directly reflecting the degree of equipment aging. It then extracts the fault records of each device from the fault history database for the past year and calculates the deduction using the formula "Historical Fault Frequency Deduction = Fault Count × 5". The higher the fault frequency, the larger the deduction value, reflecting the equipment failure risk. Finally, it obtains the temperature, humidity, and dust concentration data of the environment in which each device is located from the environmental monitoring system. According to the pre-set environmental adaptability scoring table (Excellent 100 points, Good 80 points, Average 60 points, Poor 40 points), the environmental adaptability score is determined; the response time of the most recent maintenance of each device is retrieved from the maintenance record database, and the score is calculated according to "Maintenance Response Timeliness Score = 100 - Delay Days" (full marks for maintenance completed within 24 hours of failure), reflecting the efficiency of operation and maintenance response; the inventory status of key spare parts for each device is queried from the spare parts management system, and the score is calculated according to "Spare Parts Availability Score = 100 - (Estimated Delivery Days × 2)" (full marks for all spare parts in stock), ensuring the feasibility of maintenance execution.

[0102] Subsequently, the scores of the five items were weighted and summed according to their respective weights (operational stability 0.3, historical failure frequency 0.25, environmental adaptability 0.2, maintenance response timeliness 0.15, spare parts availability 0.1) to obtain the final health status score for each device. Devices were then sorted from lowest to highest score to generate an equipment-level maintenance priority ranking table: devices with scores below 60 were classified as high-risk and an emergency maintenance work order was generated; devices with scores between 60 and 80 were classified as medium-risk and a planned maintenance work order was generated; devices with scores above 80 were classified as low-risk and their status was recorded without generating a work order.

[0103] The generated work order includes a maintenance item list (such as sensor calibration, switch contact cleaning, etc.), the name and quantity of required spare parts, a suggested execution time window (avoiding periods of high system load), a safety operation procedure number, and risk warning information (such as precautions for power outage operations). The work order is transmitted to the operation and maintenance management platform via an encrypted communication protocol, using the national cryptographic algorithm SM4 for data encryption and SM2 for digital signature to prevent tampering or leakage during transmission. After receiving the work order on a mobile device, operation and maintenance personnel must authenticate via fingerprint biometrics to view the detailed operation steps, ensuring the security and traceability of work order execution.

[0104] This preventative maintenance process transforms the traditional "post-event repair" into "pre-event planning," enabling early identification of high-risk equipment, rational allocation of maintenance resources, reduction of unplanned downtime, extension of equipment lifespan, and lower DC system operation and maintenance costs.

[0105] This invention also provides a DC system full-cycle operation and maintenance support system for implementing any of the above-described DC system full-cycle operation and maintenance support methods, including:

[0106] The multi-source status sensing module is used to collect real-time all-round operating status data of the DC system;

[0107] The data preprocessing module is used to perform time-series synchronization and normalization on the collected data to generate a standardized time-series dataset.

[0108] The fault prediction module is used to perform a probabilistic assessment of potential fault risks by calling a pre-trained fault prediction model based on the standardized time series dataset.

[0109] The isolation decision execution module is used to generate and issue the optimal fault isolation path instruction when the fault risk exceeds the threshold, and control the intelligent execution unit to complete the isolation operation.

[0110] A closed-loop feedback optimization module is used to adaptively optimize the parameters of the fault prediction module and the isolation decision execution module online based on the performance deviation of the isolated system.

[0111] The preventive maintenance planning module is used to periodically generate equipment maintenance priority rankings and digital work orders during normal system operation.

[0112] In this embodiment, the overall system architecture includes six functional modules, which work together to achieve full-cycle operation and maintenance support for the DC system. Figure 1 The multi-source status sensing module is responsible for real-time acquisition of system operation data. By deploying voltage sensors (sampling frequency 1000 times / second, resolution ≥0.1V), current sensors (sampling frequency 1000 times / second, resolution ≥0.01A), temperature sensors (sampling frequency 50 times / second, accuracy ±0.5℃), insulation monitoring devices based on the unbalanced bridge method (sampling resistance 100kΩ, accuracy 0.1 level), and intelligent terminals with built-in communication status monitoring at key nodes such as DC bus, branch outlets, and heat-sensitive parts, a multi-dimensional data stream covering the entire system topology is formed. All sensors are connected to the central acquisition unit through industrial Ethernet or fiber optic ring network to ensure that the data timestamp error is <10μs.

[0113] After receiving the raw data transmitted by the multi-source state sensing module, the data preprocessing module first uses cubic spline interpolation alignment to synchronize the time sequence of data from each channel, resamples at 10ms intervals to generate a fixed frequency sequence, and then normalizes each physical quantity to the 0-1 interval (different physical quantities correspond to their own rated ranges) using the maximum-minimum scaling method. Abnormal data is marked and stored separately, and finally a standardized time-series dataset containing multiple feature sequences such as voltage, current, and temperature is generated. Figure 3 ).

[0114] The fault prediction module is deployed on the central processing server and has a built-in pre-trained long short-term memory network model. Figure 2 After receiving the standardized time-series dataset output by the data preprocessing module, the system extracts features such as voltage sags and current surges through three hidden layers, and outputs the probability values ​​of failure of each key device in the next 10s, 30s, and 60s, providing risk warnings for the system.

[0115] The isolation decision execution module is activated when the output probability of the fault prediction module exceeds a threshold, and it has a built-in fault isolation decision engine. Figure 4 , Figure 7 Based on the system topology, load distribution, backup capacity, and equipment health score, the system generates the optimal isolation command and sends it to the intelligent execution unit (including programmable logic controller, solid-state switch array, and mechanical disconnect switch drive mechanism) to control it to complete millisecond-level electrical disconnection and physical isolation, ensuring that the faulty branch is disconnected from the system.

[0116] The closed-loop feedback optimization module starts after the isolation operation is completed. Figure 5 Once the system reaches steady state (voltage fluctuation < 0.5% for 5 seconds), parameters such as bus voltage recovery accuracy are collected. The deviation from the expected value is calculated using weighted Euclidean distance. Based on this, the weights of the fault prediction model and the parameters of the isolation decision algorithm are fine-tuned to achieve system self-optimization.

[0117] The preventative maintenance planning module is activated on a 24-hour cycle. Figure 6 Based on factors such as equipment runtime, fault records, and environmental adaptability, a health score is calculated, maintenance priority ranking and digital work orders are generated, and encrypted pushes are sent to the operation and maintenance platform to guide proactive maintenance.

[0118] Through the coordinated operation of six modules, the system achieves full-process intelligent management of DC systems, from state perception, fault prediction, isolation and handling to maintenance planning, completely changing the traditional passive operation and maintenance mode and greatly improving the system's operational reliability and maintenance efficiency.

[0119] In some embodiments, the intelligent execution unit includes a programmable logic controller, a solid-state switch array based on a wide bandgap semiconductor device, and a mechanical disconnect switch drive mechanism with high-precision position feedback. The programmable logic controller is used to coordinate and control the timing operations of the solid-state switch array and the mechanical disconnect switch drive mechanism.

[0120] In this embodiment, the intelligent execution unit in the system consists of three parts: a programmable logic controller (PLC), a solid-state switch array based on wide-bandgap semiconductor devices, and a mechanical disconnect switch drive mechanism with high-precision position feedback. These three parts, under the coordinated control of the PLC, achieve rapid isolation of faulty branches. Figure 4 ).

[0121] Solid-state switch arrays based on wide bandgap semiconductor devices are constructed using silicon carbide power devices. These devices have the characteristics of low on-resistance (<5mΩ), fast turn-off speed (<20μs), and strong impact resistance (can withstand instantaneous impact of 10 times the rated current). They can cut off the fault branch current in milliseconds, quickly block the fault propagation path, buy time for subsequent physical isolation, and prevent the fault from expanding and affecting non-faulty areas.

[0122] The mechanical disconnect switch drive mechanism adopts a combination design of permanent magnet synchronous motor and worm gear reducer. The permanent magnet synchronous motor provides stable driving force, while the worm gear reducer achieves speed reduction and torque increase, ensuring stable operation of the mechanism under extreme conditions (such as low temperature and high humidity). The mechanism has a built-in Hall position sensor that can monitor the position of the moving contact of the disconnect switch in real time, with a feedback accuracy of ±0.5°, providing a precise position signal to the programmable logic controller to ensure that the disconnect switch can accurately reach the physical isolation position. At the same time, the drive torque reserve coefficient of the mechanism is not less than 1.5, further ensuring the reliability of the operation.

[0123] As the control core of the intelligent execution unit, the programmable logic controller (PLC) is responsible for coordinating the timing of actions between the solid-state switch array and the mechanical disconnector drive mechanism. After receiving the instructions from the isolation decision execution module, it first parses the switch action sequence and time requirements in the instructions, sends a cut-off signal to the solid-state switch array, and after receiving the solid-state switch cut-off confirmation signal, it delays for 50ms before sending the action instructions to the mechanical disconnector drive mechanism to avoid conflict between the two actions. During the operation of the mechanical disconnector, it receives the position signal fed back by the Hall position sensor in real time. When it confirms that the moving contact has reached the fully disconnected position, it determines that the isolation operation is completed, generates an operation report, and uploads it to the central monitoring platform.

[0124] Through this structural design, the intelligent actuator not only utilizes solid-state switches to achieve rapid interruption of fault current, but also achieves safe and reliable physical isolation through mechanical isolating switches, balancing isolation speed and safety, ensuring foolproof fault isolation operations, and providing a guarantee for stable system operation.

[0125] In some implementations, a human-machine interface is also included to graphically display the system topology, equipment health status heatmap, fault risk prediction curve, isolation operation status and maintenance work order records in real time, and to support manual triggering of equipment self-test and temporary adjustment of isolation strategy.

[0126] In this embodiment, the system is also equipped with a human-machine interface, which is linked in real time with modules such as multi-source status perception, fault prediction, and isolation decision execution. It can display the system's operating status and operational information to maintenance personnel in an intuitive graphical manner. Figure 8 ).

[0127] The main interface displays the real-time topology of the DC system, with each device represented by an icon. The icon color dynamically changes based on the device's health status score: a score above 90 indicates normal operation; a score between 70 and 90 indicates minor risk; and a score below 70 indicates high risk requiring attention. Maintenance personnel can click on any device icon to access a detailed information window, viewing the device's real-time operating parameters (voltage, current, temperature, etc.), historical fault records, maintenance recommendations, and a fault risk prediction curve for the next 24 hours (updated every 10 minutes based on data from the fault prediction module), providing a comprehensive understanding of the device's status.

[0128] When the system performs fault isolation operations, the interface displays the isolation status in the form of dynamic animation: the branch being isolated is highlighted with flashing red to prompt maintenance personnel to pay attention to the operation progress; the branch that has been isolated is marked with a gray dashed box to clearly indicate the scope of fault isolation; the backup power supply path is displayed with a thick blue solid line to intuitively present the system power restoration path, making it easy for maintenance personnel to track the isolation operation progress in real time.

[0129] The interface also provides a historical maintenance work order query function, which displays all work order numbers, corresponding equipment names, maintenance content, executors, completion time and acceptance results in a list format. Maintenance personnel can filter and query by time, equipment type and other conditions to easily trace the maintenance history.

[0130] In addition, the interface supports manual operation by maintenance personnel: it can trigger the device self-test process, which includes sensor calibration test, communication link stress test, switch action reliability test and insulation resistance retest. After the self-test is completed, a report is automatically generated and archived. It can also temporarily adjust the priority of isolation strategy, manually specify the priority protection level of a certain type of load in the isolation decision. This adjustment is only effective for the current decision cycle and will automatically restore the default setting in the next cycle, ensuring that maintenance personnel can flexibly intervene in the system operation in extreme cases.

[0131] All manual operations are recorded, including the operator's identity, operation time, and operation content, forming a complete audit log for easy accountability. Through this human-machine interface, maintenance personnel can gain real-time and comprehensive control over the system status and flexibly intervene, improving system controllability, facilitating maintenance management, and ensuring the system's safe and stable operation even under complex conditions.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0133] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for full-cycle operation and maintenance support of a DC system, characterized in that, include: S1: Through a multi-source sensor network deployed at key nodes of the DC system, real-time electrical, thermodynamic and communication status data of the system are collected to form a multi-dimensional real-time operating status data stream covering the entire topology of the system; S2: Perform time-series synchronization and normalization processing on the multi-dimensional real-time running status data stream to generate a standardized time-series dataset containing multiple feature sequences; S3: Based on the standardized time series dataset, call the pre-trained fault prediction model to perform a probabilistic assessment of the potential fault risks of the DC system within a future preset time window, and output the fault occurrence probability value of each key device. S4: When the probability of failure of any critical equipment exceeds the preset threshold, the fault isolation decision engine is triggered to generate the optimal fault isolation path instruction based on the system topology, load distribution, backup capacity and equipment health status. S5: Send the optimal fault isolation path instruction to the corresponding intelligent execution unit to control it to complete the electrical disconnection and physical isolation operation of the fault branch; S6: After completing the fault isolation operation, collect the steady-state operating parameters of the isolated system and calculate the deviation with the expected performance indicators. Based on the calculation results, perform online adaptive optimization of the parameters of the fault prediction model and the fault isolation decision engine. S7: During normal system operation, preventive maintenance planning is periodically initiated, generating equipment-level maintenance priority ranking and digital work orders based on multi-dimensional equipment status information.

2. The method for full-cycle operation and maintenance support of a DC system according to claim 1, characterized in that, In step S2, the timing synchronization adopts an interpolation alignment method based on a unified time reference, and the normalization process adopts a maximum-minimum scaling method to compress the data of each physical quantity to a predetermined range. Data points that exceed the rated operating range are marked as abnormal data and stored separately.

3. The method for full-cycle operation and maintenance support of a DC system according to claim 1, characterized in that, In step S3, the fault prediction model is a neural network model built on a long short-term memory network. Its input is a standardized time series dataset of historical time windows, and its output is the probability value of fault occurrence at different time granularities in the future. The model is trained through supervised learning using a sample library containing time series data of historical fault events.

4. The method for full-cycle operation and maintenance support of a DC system according to claim 1, characterized in that, In step S4, the fault isolation decision engine generates the optimal fault isolation path instruction by executing the following steps: a. Identify the electrical island where the faulty device is located; b. Calculate the set of all possible isolation boundary nodes within the electrical island; c. Assess the impact of disconnecting each boundary node on the continuity of power supply to non-faulty areas and the complexity of the operation; d. Select the combination of boundary nodes with the least impact and lowest operational complexity as the isolation execution point, and generate an instruction containing the sequence of switches to be operated and a list of load transfer priorities.

5. The method for full-cycle operation and maintenance support of a DC system according to claim 1, characterized in that, In step S5, the intelligent execution unit includes a programmable logic controller, a solid-state switch array, and a mechanical disconnect switch drive mechanism. After parsing the instructions, the programmable logic controller first controls the solid-state switch array to perform millisecond-level electrical disconnection according to a preset timing sequence, and then drives the mechanical disconnect switch drive mechanism to complete physical isolation.

6. The method for full-cycle operation and maintenance support of a DC system according to claim 1, characterized in that, In step S6, the deviation calculation adopts the weighted Euclidean distance formula to comprehensively evaluate the bus voltage recovery accuracy, load power supply continuity, standby path temperature rise rate and total system loss increment. When the deviation exceeds the tolerance threshold, online fine-tuning of the weight parameters of the fault prediction model and the path optimization algorithm parameters of the isolation decision engine is triggered.

7. The method for full-cycle operation and maintenance assurance of a DC system according to claim 1, characterized in that, In step S7, the multi-dimensional status information of the equipment includes the cumulative running time of the equipment, historical fault records, environmental adaptability, maintenance response timeliness and spare parts availability; the equipment-level maintenance priority ranking is determined by calculating the equipment health status score through a multi-index comprehensive evaluation method.

8. A DC system full-cycle operation and maintenance support system, used to implement the DC system full-cycle operation and maintenance support method as described in any one of claims 1-7, characterized in that, include: The multi-source status sensing module is used to collect real-time all-round operating status data of the DC system; The data preprocessing module is used to perform time-series synchronization and normalization on the collected data to generate a standardized time-series dataset. The fault prediction module is used to perform a probabilistic assessment of potential fault risks by calling a pre-trained fault prediction model based on the standardized time series dataset. The isolation decision execution module is used to generate and issue the optimal fault isolation path instruction when the fault risk exceeds the threshold, and control the intelligent execution unit to complete the isolation operation. A closed-loop feedback optimization module is used to adaptively optimize the parameters of the fault prediction module and the isolation decision execution module online based on the performance deviation of the isolated system. The preventive maintenance planning module is used to periodically generate equipment maintenance priority rankings and digital work orders during normal system operation.

9. The DC system full-cycle operation and maintenance support system according to claim 8, characterized in that, The intelligent execution unit includes a programmable logic controller, a solid-state switch array based on wide bandgap semiconductor devices, and a mechanical disconnect switch drive mechanism with high-precision position feedback. The programmable logic controller is used to coordinate and control the timing actions of the solid-state switch array and the mechanical disconnect switch drive mechanism.

10. The DC system full-cycle operation and maintenance support system according to claim 8, characterized in that, It also includes a human-machine interface, which graphically displays the system topology, equipment health status heatmap, fault risk prediction curve, isolation operation status and maintenance work order records in real time, and supports manual triggering of equipment self-test and temporary adjustment of isolation strategy.

Images