An automatic detection platform for 10kV power distribution network intelligent terminal

Abstract

The application discloses a kind of 10kV distribution network intelligent terminal's automated detection platform, it is related to distribution network technical field, the present application includes three steps: import distribution network GIS vector data, line electrical parameter and nearly 5 years fault record, build hardware architecture, generate 1:1 digitization twin test topology and establish real-time mapping link of hardware and topology;Access FTU, DTU, TTU terminal, configure fault propagation, load fluctuation, electromagnetic interference composite test scene, collect full amount electrical parameter, communication message and instruction execution data;Through virtual terminal generation ideal data set, with real data millisecond level comparison, using deep reinforcement learning algorithm calibration system error, based on multi-modal data fusion diagnostic model analysis defect type and locate fault source.The platform improves detection precision and scene restoration degree, realizes defect accurate diagnosis, provides efficient support for distribution network terminal operation and maintenance, and guarantees the stable operation of distribution network.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and more specifically to an automated detection platform for a 10kV power distribution network intelligent terminal. Background Technology

[0002] As my country's power distribution network upgrades towards "intelligent, efficient, and reliable" systems, the 10kV distribution network, as the core link connecting the power system and users, directly impacts power supply quality and residential electricity consumption. Currently, a large number of intelligent terminals such as FTUs, DTUs, and TTUs are deployed in the 10kV distribution network. These terminals undertake key functions such as telemetry, remote signaling, and remote control, and are core equipment for achieving self-healing of distribution network faults, load regulation, and distributed power source absorption. Therefore, an automated detection platform for intelligent terminals in the 10kV distribution network is needed.

[0003] The existing technology, such as the invention patent application with publication number CN105743108B, discloses a 10kV low-voltage intelligent distribution network system with three-phase imbalance monitoring function, including a terminal part, a computer part, a communication part, and a monitoring part. The terminal part includes a feeder automation terminal, a distribution transformer detection terminal, a ring main unit, and a switching station; the computer part includes a network workstation, an application server, a front-end server, a firewall router, and a shared printer; the communication part includes an MSTP device, an Ethernet switch, and an EPON; the monitoring part includes a three-phase voltage signal processing module, a three-phase current signal processing module, a sampling circuit, and a data processing module. The data output by the data processing module is uploaded to the data bus, where the distribution management workstation receives, stores, analyzes, and processes the data.

[0004] Regarding the above solutions, the applicant of this invention has found that the above technologies have at least the following technical problems: 1. Existing technologies lack systematic quantitative logic in the calculation of the confidence level of the root cause of the fault, resulting in insufficient accuracy and traceability of the results. Most detection methods determine the confidence level only by matching the frequency of historical fault cases, without combining the abnormal data characteristics of the current terminal with the fit of typical cases for correction, resulting in the confidence level failing to truly reflect the reliability of root cause location. Although some methods introduce multi-factor adjustments, the adjustment weights mostly rely on experience settings, lacking data-driven quantitative basis, and do not dynamically optimize by associating with the terminal's historical operation and maintenance records, making it difficult to adapt to the individual differences of different brands and models of terminals. At the same time, the confidence level output only presents a single value, without indicating key evidence such as the number of cases and feature comparison results in the calculation process, making it impossible for operation and maintenance personnel to verify the rationality of the results, which can easily lead to extended maintenance cycles due to misjudgment of the root cause.

[0005] 2. Existing technologies for assessing the impact of faults are one-sided and lack an analytical system deeply integrated with the actual operation of distribution networks. Current assessments often focus on the functional status of the terminals themselves, simply determining whether a function is normal or malfunctioning, without detailing the specific manifestations and technical principles of functional degradation, thus failing to provide precise repair guidance for operation and maintenance. At the distribution network level, assessments are mostly qualitative descriptions, failing to combine terminal installation locations and distribution network topology to quantitatively analyze the impact on power supply reliability, power quality, and load distribution, making it difficult to support risk prediction for distribution network dispatching. Furthermore, there is a lack of unified grading standards for risk assessment of equipment safety, personnel safety, and grid safety, making it impossible to distinguish the degree of urgency of risks and easily leading to misallocation of operation and maintenance resources.

[0006] 3. Existing technology assessment results are disconnected from operational needs, resulting in insufficient information transmission efficiency and decision support capabilities. Most detection systems only output results in text or simple lists, failing to structure and integrate multi-dimensional information such as terminal function impact, distribution network operation risks, and security vulnerabilities. Operation and maintenance personnel must manually analyze the relationships, increasing decision-making time. While some systems include data charts, they do not generate visual diagrams based on the distribution network topology, failing to intuitively present the scope and propagation path of defects. Furthermore, assessment results are not integrated with operational processes, lacking clear priorities and pre-treatment recommendations for different levels of impact, leading to a disconnect between assessment and execution. This hinders rapid conversion into maintenance actions and prolongs the impact time of terminal failures on distribution network operation. Summary of the Invention

[0007] To address the aforementioned technical shortcomings, the present invention aims to provide an automated detection platform for a 10kV distribution network intelligent terminal.

[0008] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides an automated detection platform for a 10kV distribution network intelligent terminal, including the following steps: Step 1, Construction of distribution network twin detection environment: Import the GIS vector data, line electrical parameters and complete historical fault records of the target 10kV distribution network in the past 5 years, build a hardware architecture consisting of a main control unit, distributed detection nodes, dynamic load simulation module and temperature electromagnetic composite environment box, generate a 1:1 digital twin test topology based on the actual wiring relationship of the distribution network, and establish a real-time mapping link between the hardware module signal interface and the twin topology node.

[0009] Step 2, Multi-terminal Topology Linkage Testing: Connect the FTU, DTU, and TTU to be tested to the corresponding distributed testing nodes according to their actual distribution network installation locations. Based on the twin topology configuration, a composite test scenario including dynamic fault propagation, random load fluctuations, and multi-band electromagnetic interference is set up to trigger the collaborative control response between terminals and collect all electrical parameters, protocol communication messages, and control command execution data during the operation of the terminals.

[0010] Step 3: Virtual-Real Closed-Loop Self-Calibration and Defect Diagnosis: Start the virtual terminal to generate an ideal response dataset that conforms to the distribution network operation standard, and compare it with the real-time detection data of the real terminal at the millisecond level. Use deep reinforcement learning algorithm to dynamically adjust the hardware test signal parameters to compensate for system errors. At the same time, analyze the terminal defect type and locate the root cause of the fault based on the multimodal data fusion diagnostic model.

[0011] The beneficial effects of this invention are as follows: 1. In the embodiments of this invention, a high-fidelity distribution network twin testing environment is constructed, realizing a leap from single-condition testing to full-scenario simulation verification, significantly improving the consistency between testing and actual operation. The platform imports distribution network GIS vector data, line electrical parameters, and historical fault records from the past 5 years, generating a 1:1 digital twin test topology based on the Unity3D engine, accurately restoring the actual wiring relationships and equipment characteristics of the distribution network; simultaneously, combined with a temperature-electromagnetic composite environment box and a dynamic load simulation module, it can synchronously reproduce complex scenarios such as dynamic fault propagation, random load fluctuations, and multi-band electromagnetic interference superposition, rather than the traditional fixed single test condition. This high-fidelity environment can effectively simulate various complex conditions of the terminal in actual operation, avoiding the disconnect between laboratory testing and field operation, ensuring that the test results can directly reflect the terminal's operating performance in a real distribution network environment, and providing a more accurate basis for terminal reliability assessment.

[0012] 2. This invention, through its innovative virtual-real closed-loop self-calibration mechanism and full-data acquisition scheme, significantly improves the accuracy of detection data and system stability. The platform generates an ideal response dataset conforming to power distribution network operation standards via a virtual terminal, comparing it with real terminal detection data at millisecond levels. Combined with deep reinforcement learning algorithms, it dynamically adjusts hardware test signal parameters, controlling the detection error of key parameters such as voltage and current to within 0.1%, far exceeding the error level of traditional detection systems. Simultaneously, distributed detection nodes incorporate high-precision acquisition modules, acquiring full electrical parameters, protocol communication messages, and control command execution data at intervals ≤100μs. This, coupled with 5G edge computing nodes, enables real-time data transmission and encrypted storage, ensuring data integrity and timeliness. This precise acquisition + dynamic calibration model not only solves the problems of low accuracy and susceptibility to interference in traditional detection data but also provides a high-quality data foundation for subsequent defect diagnosis, guaranteeing the reliability of the detection results.

[0013] 3. This invention's embodiment features a multi-dimensional integrated defect diagnosis system, upgrading from fuzzy judgment to precise location, significantly improving defect identification efficiency and root cause tracing capabilities. The platform constructs a multi-modal data fusion diagnostic model based on graph neural networks. Through an attention mechanism, it rationally allocates weights for electrical parameters, communication messages, and control behavior data, accurately distinguishing between three defect types: hardware faults, software vulnerabilities, and protocol incompatibility. Simultaneously, a defect-root cause association database containing over 10,000 cases is introduced. This database, combined with terminal hardware model, software version, and historical maintenance records, locates the root cause of the fault. The root cause confidence is calculated using three dimensions: basic confidence, feature fit, and historical maintenance weights, ensuring the reliability of root cause location. Compared to traditional detection methods that can only determine the existence of a defect but not its root cause, this system can improve defect diagnosis accuracy to over 99% and reduce root cause location time to the second level. This provides clear guidance for maintenance personnel to quickly develop repair plans, reducing the time and cost waste caused by blind repairs.

[0014] 4. This invention provides a systematic fault impact assessment and visualization, constructing a closed-loop support system for detection, diagnosis, and operation and maintenance, significantly improving the efficiency and safety of distribution network operation and maintenance decisions. The platform conducts impact assessments from three dimensions: terminal function, distribution network operation, and security risks. It employs a three-level judgment standard to refine the degree of functional impact, quantifies the impact on power supply reliability and power quality, and clarifies the risk levels of equipment, personnel, and the power grid. Simultaneously, the assessment results are presented in structured tables and visual diagrams. The visual diagrams can mark the affected areas in the distribution network topology, intuitively displaying the scope and propagation path of the defect, and linking it to operation and maintenance priorities and pre-treatment suggestions. This assessment model not only solves the problems of one-sided dimensions and fragmented information in traditional assessments but also directly connects to the operation and maintenance process, helping operation and maintenance personnel quickly determine the urgency of defects and formulate targeted measures, such as emergency repairs or planned maintenance, effectively reducing the impact of terminal faults on distribution network operation, reducing power outage duration and safety hazards, and providing strong support for the stable operation of the distribution network. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Examples of embodiments of the present invention Figure 1 As shown, an automated testing platform for a 10kV distribution network intelligent terminal includes the following steps: Step 1: Construction of the distribution network twin testing environment: Import the GIS vector data, line electrical parameters and complete historical fault records of the target 10kV distribution network in the past 5 years, build a hardware architecture consisting of a main control unit, distributed testing nodes, dynamic load simulation modules and temperature-electromagnetic composite environment box, generate a 1:1 digital twin test topology based on the actual wiring relationship of the distribution network, and establish a real-time mapping link between the hardware module signal interface and the twin topology nodes.

[0019] In a specific embodiment, the import process of the target 10kV distribution network's GIS vector data, line electrical parameters, and complete historical fault records for the past 5 years is as follows: The GIS vector data includes the distribution network line coordinates, tower locations and models, transformer capacity, location, switchgear models, and installation nodes; the line electrical parameters include line resistance, line inductance, line capacitance, and line insulation resistance; the complete historical fault records for the past 5 years include the fault occurrence time, fault type, fault occurrence line segment and tower number, fault duration, fault handling measures, and recovery time. The data format supports importing mainstream formats such as SHP, Excel, and JSON.

[0020] It should be noted that the accuracy of the distribution network line coordinates in the GIS vector data must be ≤1m, and the tower position coordinate error must not exceed 0.5m. Equipment model information should include the manufacturer, serial number, and commissioning time to ensure that the equipment parameters are consistent with the actual situation during topology modeling. For line electrical parameters, the resistance, inductance, and capacitance should provide values ​​per kilometer. The insulation resistance of the line should be marked with the test environment temperature and humidity to avoid parameter deviations due to environmental differences. The complete historical fault records for the past 5 years should include fault occurrence times accurate to the second, and fault handling measures should include maintenance personnel, equipment used, and repair processes. When importing data formats, SHP format should support automatic conversion between projected coordinate systems and geographic coordinate systems, Excel format should have a preset standardized template, and JSON format should support nested data parsing. At the same time, a data integrity verification step should be added, such as missing field prompts and format error markings. If there are parameter anomalies in the imported data, such as line resistance exceeding the normal range, the system will automatically trigger an anomaly warning and provide a reference correction value to ensure that the imported data meets the high-precision requirements of twin topology construction and test scenario configuration.

[0021] In a specific embodiment, the hardware architecture consisting of a main control unit, distributed detection nodes, a dynamic load simulation module, and a temperature-electromagnetic composite environment box is constructed as follows: The hardware architecture includes one industrial-grade main control unit, 3-12 distributed detection nodes, each node equipped with 6 standard interfaces supporting USB3.0, RS485, Ethernet, and aviation plug interface types, with a maximum power supply of ≤100W per node, a dynamic load simulation module, and a temperature-electromagnetic composite environment box; each distributed detection node is connected to the main control unit via single-mode fiber optic Ethernet; the main control unit integrates a wide-range adaptive signal source with a voltage output range of 0-12kV, a current output range of 0-200A, a frequency adjustment range of 45-55Hz, and an adjustment step of 0.01Hz.

[0022] It should be noted that the industrial-grade main control unit uses a quad-core or higher processor with a main frequency of ≥2.4GHz, memory ≥16GB, storage capacity ≥1TB, and an operating temperature range of -20℃ to 60℃. It has triple hardware protection functions for overvoltage, overcurrent, and overtemperature to ensure stable operation under complex testing scenarios. Each interface of the distributed detection node must be labeled with its rated voltage and signal transmission rate. The interface is equipped with anti-misinsertion clips and electrostatic protection components, and each node has a built-in independent cooling fan to prevent overheating during long-term operation. The dynamic load simulation module must support independent adjustment of resistive, inductive, and capacitive loads, with resistive load ranges of 0.1Ω-10kΩ, inductive load ranges of 1mH-100mH, and capacitive load ranges of 1μF-100μF, with a load adjustment accuracy of ±1%, and a load change response function to simulate the instantaneous characteristics of power distribution network load fluctuations. The temperature-electromagnetic composite environmental chamber must specify a temperature control accuracy of ±0. The test chamber must be set at 5℃ with a humidity control range of 20%-90%RH. Electromagnetic interference must support both conducted and radiated interference modes, with conducted interference frequencies of 150kHz-30MHz and radiated interference field strength of 0-100V / m. An insulated load-bearing tray must be installed inside the chamber to prevent leakage during terminal testing. The single-mode fiber optic Ethernet connection between each distributed detection node and the main control unit must use an LC-type interface, with a fiber optic transmission distance ≤2km. The connection link must be shielded to reduce the impact of electromagnetic interference on data transmission. The wide-range adaptive signal source integrated into the main control unit must be labeled with output accuracy and ripple coefficient ≤0.5%, and possess a customizable signal waveform function. Output parameters can be adjusted in real-time via the main control unit software. Simultaneously, a synchronous control link must be established between the signal source, the dynamic load simulation module, and the temperature-electromagnetic composite environment chamber via an RS485 bus to ensure coordinated triggering of signal output, load changes, and environmental interference in the test scenario.

[0023] In one specific embodiment, the generation of a 1:1 digital twin test topology based on the actual wiring relationship of the distribution network, and the establishment of a real-time mapping link between the hardware module signal interface and the twin topology nodes, are carried out as follows: A 3D model of the twin topology is constructed using the Unity3D engine, and line nodes and equipment models are associated according to the actual wiring relationship of the distribution network. The electrical parameter attributes of each node are set to be consistent with the imported line electrical parameters. Each interface of the distributed detection node is bound to the access node of the corresponding terminal in the twin topology. The voltage output terminal of the wide-area adaptive signal source is mapped to the line voltage acquisition node of the twin topology, the current output terminal is mapped to the line current acquisition node, and the dynamic load simulation module is mapped to the load node of the twin topology. The mapping link adopts a heartbeat detection mechanism, which checks the signal transmission status every 500ms. If the link is interrupted, the main control unit automatically triggers an alarm and attempts to reconnect, with a reconnection time of ≤3s.

[0024] It should be noted that when building the twin topology 3D model using the Unity3D engine, the line model must have its appearance texture and physical properties set according to the actual conductor material. The scale accuracy of the tower and equipment models must be ≤1%. Key components, such as transformer bushings and switch contacts, should be finely modeled, and the model must support LOD rendering technology to improve running smoothness while ensuring visualization effects. When setting the electrical parameter attributes of each node, a parameter association table must be established to directly associate the imported line resistance, inductance, capacitance, and other parameters with the corresponding line segment's electrical attribute module in the twin topology. Parameter modifications must support real-time synchronization to avoid test deviations caused by parameter inconsistencies. When binding the hardware interface to the twin topology nodes, bidirectional verification must be performed: on the one hand, a test signal is sent through the main control unit to check whether the corresponding node in the twin topology accurately receives and responds to the signal strength; on the other hand, terminal action commands are simulated in the twin topology. The system verifies whether the hardware interface can accurately trigger the response of the real terminal. In addition to checking the signal transmission status every 500ms, the heartbeat detection mechanism also needs to define specific verification indicators. If a link interruption occurs, the main control unit must issue an audible and visual alarm with a sound level ≥80dB and a red flashing alarm light to indicate the fault. Simultaneously, the interruption time, interruption node, and fault code should be recorded. During reconnection, automatic recovery should be prioritized, such as restarting the corresponding detection node or resetting the signal interface. If reconnection fails within 3 seconds, it should automatically switch to a backup link, such as a backup fiber optic interface or a 4G backup communication module. Furthermore, the twin topology must support dynamic updates. When the actual wiring of the distribution network changes, such as adding lines or replacing equipment, the topology structure can be automatically adjusted by importing updated GIS data. Historical topology versions must be retained during the update process for easy traceability and rollback, while ensuring that the hardware mapping link is not interrupted during topology updates and does not affect ongoing detection tasks.

[0025] Step 2, Multi-terminal Topology Linkage Testing: Connect the FTU, DTU, and TTU to be tested to the corresponding distributed testing nodes according to their actual distribution network installation locations. Based on the twin topology configuration, a composite test scenario including dynamic fault propagation, random load fluctuations, and multi-band electromagnetic interference is set up to trigger the collaborative control response between terminals and collect all electrical parameters, protocol communication messages, and control command execution data during the operation of the terminals.

[0026] In a specific embodiment, the configuration of the composite test scenario based on twin topology, including dynamic fault propagation, random load fluctuation, and multi-band electromagnetic interference, is as follows: Dynamic fault propagation scenario: An initial fault is set at any node of the twin topology backbone, the fault propagation speed is set to 0.5-2km / s, and adjusted in steps of 0.1km / s. The fault type can be switched to two-phase short circuit and three-phase short circuit. At the same time, the fault clearing time is configured to 0.1-5s, with an adjustment step of 0.01s, and the hardware synchronously outputs the corresponding fault current and voltage signals.

[0027] Random load fluctuation scenario: Set the load fluctuation mode in one or more load nodes of the twin topology. The base load value is 50%-120% of the terminal rated load, the fluctuation frequency is 0.1-1Hz, the adjustment step is 0.01Hz, the fluctuation amplitude is ±5%-±30%, and the adjustment step is ±1%. The dynamic load simulation module adjusts the load value in real time following the load fluctuation curve.

[0028] Multi-band electromagnetic interference scenario: Electromagnetic interference is applied through a temperature-electromagnetic composite environment box. The interference frequency bands are divided into three bands: 30MHz-100MHz, 100MHz-500MHz, and 500MHz-1GHz. The interference intensity of each band can be adjusted independently, from 0 to 100V / m, with an adjustment step of 1V / m. The interference mode supports continuous interference and pulse interference, with a pulse period of 1-10s and a pulse width of 0.1-1s.

[0029] The three scenarios can be started individually or in combination. When starting in combination, the scenario trigger interval is ≤100ms. The duration of a single test for each scenario is 5-30 minutes, set in 1-minute increments.

[0030] It should be noted that in the dynamic fault propagation scenario, the initial fault resistance value must be clearly defined: 10-100Ω for single-phase grounding faults and 0.1-10Ω for phase-to-phase short-circuit faults, with an adjustment step of 0.1Ω. Fault propagation must simulate the attenuation characteristics of different line materials, synchronously outputting transient waveforms of fault current and voltage, such as the peak multiple of short-circuit current and voltage drop amplitude. For the random load fluctuation scenario, multiple fluctuation curve types must be provided, including sine wave fluctuations, square wave fluctuations, and random noise fluctuations. Each curve must support custom parameters, such as the phase offset of sine wave fluctuations and the duty cycle of square wave fluctuations. When adjusting the load, the dynamic load simulation module must ensure that the current change rate is ≤10A / ms to avoid damage to the terminal from inrush current. In the multi-band electromagnetic interference scenario, the modulation method of the interference signal must be clearly defined; pulse interference requires… Mark the rise and fall times to ensure that the interference simulation conforms to the actual electromagnetic environment of the distribution network; when starting a scenario combination, priority rules must be clearly defined, such as the fault dynamic propagation scenario having higher priority than the load fluctuation scenario, to ensure that the fault signal triggers the terminal response first, and the combined scenario must support timing arrangement, such as starting the load fluctuation for 10 minutes first, and then triggering the fault dynamic propagation and electromagnetic interference superposition; during the test of each scenario, key data must be recorded at 1ms intervals, such as the instantaneous values ​​of current and voltage in the fault scenario, the power change curve of the load scenario, and the field strength value of the interference scenario. The data record must be associated with the twin topology node ID and the hardware module number to facilitate the subsequent traceability of the correspondence between scenario configuration and terminal response. At the same time, after the test time is reached, the scenario configuration parameters, such as fault type, fluctuation frequency and interference frequency band, must be automatically saved.

[0031] In a specific embodiment, the acquisition process of the full electrical parameters, protocol communication messages and control command execution data during the operation of the acquisition terminal is as follows: Full electrical parameter acquisition: The high-precision acquisition module built into the distributed detection node acquires the terminal input and output voltage, current, power and power factor, with an acquisition interval of ≤100μs.

[0032] Protocol communication message acquisition: The message capture module collects communication messages between the terminal and the twin topology, supporting mainstream distribution network communication protocols such as IEC61850, DL / T645, and Modbus, and records message sending and receiving timestamps, message length, and message content.

[0033] Control command execution data acquisition: The time when the terminal receives the control command, the start time of command execution, and the completion time of command execution are recorded through the timestamp module. The command response delay is calculated as follows: command response delay = execution start time - command reception time, command execution duration = execution completion time - execution start time. At the same time, terminal state change data during command execution are collected.

[0034] All collected data is transmitted to the main control unit in real time via 5G edge computing nodes, stored in an encrypted format, with a storage period of ≥3 years, and supports data interruption resume.

[0035] Step 3: Virtual-Real Closed-Loop Self-Calibration and Defect Diagnosis: Start the virtual terminal to generate an ideal response dataset that conforms to the distribution network operation standard, and compare it with the real-time detection data of the real terminal at the millisecond level. Use deep reinforcement learning algorithm to dynamically adjust the hardware test signal parameters to compensate for system errors. At the same time, analyze the terminal defect type and locate the root cause of the fault based on the multimodal data fusion diagnostic model.

[0036] In a specific embodiment, the deep reinforcement learning algorithm is used to dynamically adjust the hardware test signal parameters to compensate for system errors. The specific adjustment process is as follows: Ideal response dataset generation: Based on the distribution network operation standard and twin topology parameters, the virtual terminal generates ideal voltage waveforms, current waveforms, communication message timing and control command response time datasets for each test scenario. The data sampling interval is consistent with the real terminal detection data.

[0037] Error calculation: Align the real-time detection data of the actual terminal with the ideal response dataset by timestamp, and calculate the voltage error: Voltage error = |actual voltage value - ideal voltage value| / ideal voltage value × 100%. The current error is calculated in the same way as the voltage error. Communication message delay error =actual message transmission time - ideal message transmission time. Control command response error =actual response delay - ideal response delay. When any error > 0.1%, parameter adjustment is triggered.

[0038] Deep reinforcement learning adjustment: With all errors ≤0.1% as the optimization objective, a deep reinforcement learning model is constructed, using the DQN algorithm, with a neural network of ≥5 layers and ≥128 hidden layer nodes. The current error value is used as the state input, and the output is the voltage adjustment, current adjustment, and frequency adjustment of the wide-range adaptive signal source.

[0039] Adjustment and verification: The hardware modifies the test signal parameters according to the output adjustment amount, re-acquires terminal detection data and calculates the error. If the error is ≤0.1%, the calibration is completed; if the error is still >0.1%, the above adjustment process is repeated. The maximum number of calibrations in a single test is ≤10, and the total calibration time is ≤1s.

[0040] It should be noted that when generating the ideal response dataset, the specific distribution network operation standard is DL / T1400-2015 "Remote Terminal for Distribution Network Automation System". Simulation calculations are performed using parameters such as line impedance and load characteristics in the twin topology. The dataset includes both steady-state and transient data. Steady-state data represents continuous waveforms under rated operating conditions, while transient data represents abrupt changes in waveform at the moment of a fault. Data sampling must use a synchronous clock to ensure complete alignment with the timestamps of the actual terminal detection data. A weighting coefficient is introduced in the error calculation stage. Key electrical parameters such as voltage and current are assigned a weight of 0.6, while data such as communication message delay and control command response are assigned a weight of 0.4. The weighted total error is used to determine whether an adjustment is triggered. The weighted total error = 0.6 × (voltage error + current error) / 2 + 0.4 × (communication message delay error + control command response error) / 2, avoiding unnecessary adjustments triggered by errors in a single non-critical parameter. When training reinforcement learning models, a combination of offline pre-training and online fine-tuning is required. Offline pre-training samples consist of over 100,000 sets of historical calibration data from different terminal models and test scenarios. Online fine-tuning continuously optimizes model parameters based on real-time detection data. The voltage adjustment of the model output must be limited to the range of -0.5kV to +0.5kV, the current adjustment to the range of -5A to +5A, and the frequency adjustment to the range of -0.5Hz to +0.5Hz to prevent damage to hardware or terminals due to excessive adjustment. When verifying parameter adjustments, a repeatability verification step must be added. The weighted total error must be ≤0.1% after three consecutive calibrations to be considered a qualified calibration. If the standard is not met even after reaching the maximum number of calibrations in a single test, the main control unit must automatically record the current error data, adjust the parameter trajectory, and trigger an alarm, while also outputting hardware module fault diagnosis suggestions. In addition, after calibration, the model parameters and adjustment strategies for this calibration must be saved to establish a calibration strategy library.

[0041] In a specific embodiment, the analysis of terminal defect types based on the multimodal data fusion diagnostic model is carried out as follows: Data preprocessing: The collected abnormal electrical parameter data, abnormal communication message data, and abnormal control behavior data are standardized, the data are mapped to the 0-1 range, noise data is removed, and feature parameters such as electrical parameter fluctuation amplitude, message loss rate, and instruction response delay time are extracted.

[0042] Defect type diagnosis: The multimodal data fusion diagnostic model adopts a graph neural network architecture. The preprocessed multimodal feature data is input into the model, and the model assigns weights to each modality of data through an attention mechanism. The weights for electrical parameters are 0.4, communication messages are 0.3, and control behaviors are 0.3. The model outputs the terminal defect type, including hardware failure, software vulnerability, and protocol incompatibility.

[0043] Fault Root Cause Location: For the diagnosed defect type, the model's built-in defect-root cause association database is retrieved. The defect-root cause association database contains 10,000+ terminal fault cases, covering common fault root causes of terminals of different brands and models. Combined with the hardware model, software version and historical operation and maintenance records of the terminal under test, the specific fault root cause is located. At the same time, the confidence level of the root cause and the assessment of the fault impact range are output.

[0044] It should be noted that the synchronization clock uses BeiDou / GPS dual-mode time synchronization technology to ensure that the timestamp deviation between the ideal response dataset and the real terminal detection data is controlled within microseconds, avoiding error calculation distortion due to time asynchrony. The trigger adjustment threshold for the weighted total error needs to be explicitly set to >0.1%, and supplementary rules for abnormal data removal in error calculation should be added, such as using the 3σ criterion to remove pulse interference data during the electrical parameter acquisition process, to improve the accuracy of error calculation. In the offline pre-training stage, the Adam optimizer should be explicitly used, with a learning rate of 0.001, a weight decay coefficient of 0.0001, and ≥5000 training iterations. An early stopping strategy should be adopted during training, stopping training if the validation set loss does not decrease for 50 consecutive iterations to prevent model overfitting. In the online fine-tuning stage, the fine-tuning frequency should be set to update the model parameters once every 100 sets of real-time data collected to ensure that the model adapts to the characteristics of different terminals. The voltage, current, and frequency adjustments in the model output are further limited in step size: voltage adjustment step size ≤ 0.01kV, current adjustment step size ≤ 0.01A, and frequency adjustment step size ≤ 0.01Hz, to avoid sudden parameter changes impacting the hardware and terminal. When the main control unit triggers an alarm, the alarm format is an audible and visual alarm plus a push notification from the maintenance system. The alarm sound level is ≥ 80dB, and the alarm light uses a red flashing mode. Simultaneously, hardware faults in the output are investigated down to specific modules, such as checking the filter capacitor of the wide-range adaptive signal source and the acquisition chip of the distributed detection node. The calibration strategy library needs to be categorized and managed according to terminal model and test scenario, supporting the addition, deletion, modification, and query of strategies and version iteration. When saving a strategy, it needs to be associated with the terminal's serial number, test date, and calibration personnel information. The strategy library also needs to have access permissions set, allowing administrators to modify it and ordinary operators to only call it. It also supports exporting strategies to a standardized format for easy cross-platform sharing and reuse.

[0045] In a specific embodiment, the output of the root cause confidence and fault impact range assessment is specifically output as follows: Basic confidence acquisition: retrieve historical fault cases that match the current defect type, terminal hardware model, and software version from the defect-root cause association database, and count the frequency of occurrence of the same fault root cause in the matching cases. Basic confidence = (frequency of occurrence of the same fault root cause / total number of matching cases) × 100%.

[0046] Feature fit correction: Compare the electrical parameter fluctuation range, message loss rate and instruction response delay time corresponding to the abnormal data features of the current terminal with the typical data features corresponding to the fault root cause in the matching case, calculate the feature fit, and the corrected confidence = base confidence × (0.6 + 0.4 × feature fit).

[0047] Historical maintenance weight adjustment: If the terminal to be tested has historical maintenance records, when the same defect has occurred in the historical maintenance and the root cause of the repair is consistent with the current root cause, the confidence level is increased by 5%-10%; if the number of successful repairs in the past is ≥3, the confidence level is increased by 10%; if the number of successful repairs in the past is 1-2, the confidence level is increased by 5%. If the same defect in the historical maintenance corresponds to different root causes, the confidence level is reduced by 3%-8%; if there are ≥2 records of different root causes, the confidence level is reduced by 8%; if there is 1 record, the confidence level is reduced by 3%.

[0048] Final confidence score output: The final confidence score is rounded to one decimal place. The report will also indicate the basis for the confidence score calculation, including the number of matching cases, the feature fit value, and historical operation and maintenance adjustment instructions.

[0049] It should be noted that the feature fit calculation uses a cosine similarity algorithm. After standardizing the three feature data points of the current terminal—electrical parameter fluctuation amplitude, message loss rate, and command response delay duration—to the 0-1 range, a similarity calculation is performed with the typical feature vectors of the corresponding fault root causes in the defect-root cause association database. The feature fit value ranges from 0 to 1; a higher value indicates a higher degree of matching between the current abnormal feature and the typical feature. A confidence limit constraint is set for historical maintenance weight adjustments; the final confidence level after adjustment must not exceed 100%. If the adjusted confidence level exceeds 100%, 100% is automatically taken as the final value. Additionally, the effective time of historical maintenance records is also considered. The scope is the past 3 years; maintenance records older than 3 years are not included in the weight adjustment. After the confidence score is output, a reasonableness verification step is added. If the final confidence score is lower than 50%, the main control unit automatically triggers a secondary diagnostic process to retrieve more historical cases to expand the matching range and supplement the terminal with more dimensions of operational data to participate in the feature matching calculation, thereby improving the diagnostic accuracy in low-confidence scenarios. The calculation basis indicated in the report is presented in a standardized table format, which includes the total number of matching cases, the number of cases with the same fault root cause, the specific value of feature matching, the number of historical maintenance records and the adjustment range, and also indicates the algorithm name and parameter settings used in each calculation.

[0050] In a specific embodiment, the analysis and location of the fault root cause based on the multimodal data fusion diagnostic model is generated as follows: Terminal function impact assessment: Based on the defect type and fault root cause, the degree of impact on the core functions of the terminal is analyzed item by item, and a three-level judgment standard of normal, degraded and failure is adopted.

[0051] Impact assessment of distribution network operation: Based on the distribution network topology and terminal installation location, assess the impact of defects on distribution network operation indicators, including power supply reliability, power quality and load distribution.

[0052] Safety risk impact assessment: Identify potential safety hazards caused by defects, including equipment safety, personnel safety and power grid safety, and classify them into high, medium and low risk levels.

[0053] Assessment results output: The assessment results are presented in a structured table format, which includes four columns: impact dimension, specific impact description, impact degree and risk level, and estimated impact scope and consequences. Visual diagrams are also included.

[0054] It should be noted that the terminal function impact assessment uses the criteria for determining normal, degraded, and failed. Normal means that all indicators of the terminal's core functions meet the requirements of the DL / T1400-2015 standard; degraded means that the functional indicators exceed the standard's allowable range but can still maintain basic operation; failed means that the function is completely lost or the indicators are severely exceeded. The distribution network operation impact assessment quantifies the degree of impact on various operational indicators. Power supply reliability requires estimating the average outage time of the line caused by the defect; power quality is defined by the specific numerical range of voltage and frequency deviations; load allocation assessment assesses the load transfer ratio or overload risk caused by the defect, such as branch line load transfer to the main line causing the main line load rate to exceed 80%; the safety risk impact assessment sets three risk level thresholds, with high risk indicating that the defect may... Defects can cause serious accidents such as equipment burnout, power grid tripping, and electric shock. Medium-risk defects may shorten equipment lifespan and reduce power quality but pose no direct safety threat. Low-risk defects only affect local terminal functions and have no obvious safety hazards. The visualization of the assessment results must be clearly presented, including a distribution network topology map, using different colors to mark the terminal location and affected area in the twin topology (red for high risk, yellow for medium risk, and blue for low risk), and a functional impact radar chart to intuitively show the degree of impact on various core terminal functions, risk trend prediction curves, and predictions of risk development trends when defects are not addressed based on historical data. In addition, a structured table should be provided with a disposal suggestion column, giving specific operation and maintenance measures for different levels of impact and risk.

[0055] The examples described in this invention are not limited to the specific embodiments listed above. The examples are merely illustrative to facilitate understanding of the invention and do not constitute a limitation on the scope of protection of this invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of this invention should be included within the scope of protection.

[0056] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. An automated detection platform for a 10kV distribution network intelligent terminal, characterized in that, Includes the following steps: Step 1: Construction of Distribution Network Twin Detection Environment: Import the GIS vector data, line electrical parameters and complete historical fault records of the target 10kV distribution network in the past 5 years, build a hardware architecture consisting of a main control unit, distributed detection nodes, dynamic load simulation module and temperature-electromagnetic composite environment box, generate a 1:1 digital twin test topology based on the actual wiring relationship of the distribution network, and establish a real-time mapping link between the hardware module signal interface and the twin topology node. Step 2, Multi-terminal topology linkage detection: Connect the FTU, DTU and TTU to be tested to the corresponding distributed detection nodes according to the actual installation location of the distribution network. Based on the twin topology configuration, a composite test scenario including fault dynamic propagation, random load fluctuation and multi-band electromagnetic interference is set up to trigger the coordinated control response between terminals and collect all electrical parameters, protocol communication messages and control command execution data during the operation of the terminals. Step 3: Virtual-Real Closed-Loop Self-Calibration and Defect Diagnosis: Start the virtual terminal to generate an ideal response dataset that conforms to the distribution network operation standard, and compare it with the real-time detection data of the real terminal at the millisecond level. Use deep reinforcement learning algorithm to dynamically adjust the hardware test signal parameters to compensate for system errors. At the same time, analyze the terminal defect type and locate the root cause of the fault based on the multimodal data fusion diagnostic model.

2. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 1, characterized in that, The import process for the target 10kV distribution network's GIS vector data, line electrical parameters, and complete historical fault records for the past 5 years is as follows: The GIS vector data includes the coordinates of the distribution network line route, the location and model of the poles and towers, the capacity and location of the transformers, the model and installation nodes of the switchgear; the electrical parameters of the lines include the line resistance, the line inductance, the line capacitance and the line insulation resistance; the complete historical fault records for the past 5 years include the fault occurrence time, fault type, the line segment and pole number where the fault occurred, the duration of the fault, the fault handling measures and the recovery time, and the data format supports importing mainstream formats such as SHP, Excel and JSON.

3. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 2, characterized in that, The hardware architecture consisting of a main control unit, distributed detection nodes, a dynamic load simulation module, and a temperature-electromagnetic composite environment box is constructed as follows: The hardware architecture includes one industrial-grade main control unit, 3-12 distributed detection nodes, each node is equipped with 6 standard interfaces, supporting USB3.0, RS485, Ethernet and aviation plug interface types, with a maximum power supply of ≤100W per node, dynamic load simulation module, and temperature and electromagnetic composite environment box; each distributed detection node is connected to the main control unit via single-mode fiber optic Ethernet; the main control unit integrates a wide-range adaptive signal source, with a voltage output range of 0-12kV, a current output range of 0-200A, a frequency adjustment range of 45-55Hz, and an adjustment step of 0.01Hz.

4. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 3, characterized in that, The method involves generating a 1:1 digital twin test topology based on the actual wiring relationship of the distribution network, and establishing a real-time mapping link between the hardware module signal interface and the twin topology nodes. The specific mapping process is as follows: A 3D twin topology model is constructed using the Unity3D engine. Line nodes and equipment models are associated according to the actual wiring relationship of the distribution network. The electrical parameter attributes of each node are set to be consistent with the imported line electrical parameters. Each interface of the distributed detection node is bound to the access node of the corresponding terminal in the twin topology. The voltage output terminal of the wide-area adaptive signal source is mapped to the line voltage acquisition node of the twin topology, the current output terminal is mapped to the line current acquisition node, and the dynamic load simulation module is mapped to the load node of the twin topology. The mapping link uses a heartbeat detection mechanism to check the signal transmission status every 500ms. If the link is interrupted, the main control unit will automatically trigger an alarm and attempt to reconnect. The reconnection time is ≤3s.

5. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 4, characterized in that, The specific configuration process for the composite test scenario based on twin topology, which includes dynamic fault propagation, random load fluctuations, and multi-band electromagnetic interference, is as follows: Fault dynamic propagation scenario: Set an initial fault at any node of the twin topology backbone, set the fault propagation speed to 0.5-2km / s, and adjust it in 0.1km / s increments. The fault type can be switched to two-phase short circuit and three-phase short circuit. At the same time, configure the fault clearing time to 0.1-5s, with an adjustment step of 0.01s. The hardware synchronously outputs the corresponding fault current and voltage signals. Random load fluctuation scenario: Set the load fluctuation mode in one or more load nodes of the twin topology. The base load value is 50%-120% of the terminal rated load, the fluctuation frequency is 0.1-1Hz, the adjustment step is 0.01Hz, the fluctuation amplitude is ±5%-±30%, and the adjustment step is ±1%. The dynamic load simulation module adjusts the load value in real time following the load fluctuation curve. Multi-band electromagnetic interference scenario: Electromagnetic interference is applied through a temperature-electromagnetic composite environment box. The interference frequency bands are divided into three bands: 30MHz-100MHz, 100MHz-500MHz, and 500MHz-1GHz. The interference intensity of each band can be adjusted independently, from 0 to 100V / m, with an adjustment step of 1V / m. The interference modes support continuous interference and pulse interference, with a pulse period of 1-10s and a pulse width of 0.1-1s. The three scenarios can be started individually or in combination. When starting in combination, the scenario trigger interval is ≤100ms. The duration of a single test for each scenario is 5-30 minutes, set in 1-minute increments.

6. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 5, characterized in that, The acquisition process involves collecting all electrical parameters, protocol communication messages, and control command execution data during the operation of the acquisition terminal. The specific acquisition process is as follows: Full electrical parameter acquisition: The high-precision acquisition module built into the distributed detection node acquires the terminal input and output voltage, current, power and power factor, with an acquisition interval of ≤100μs; Protocol communication message acquisition: The message capture module collects communication messages between the terminal and the twin topology, supporting mainstream distribution network communication protocols such as IEC61850, DL / T645, and Modbus, and records message sending and receiving timestamps, message length, and message content. Control command execution data acquisition: The time when the terminal receives the control command, the start time of command execution, and the completion time of command execution are recorded through the timestamp module. The command response delay is calculated as follows: command response delay = execution start time - command reception time, command execution duration = execution completion time - execution start time. At the same time, terminal state change data during command execution are collected. All collected data is transmitted to the main control unit in real time via 5G edge computing nodes, stored in an encrypted format, with a storage period of ≥3 years, and supports data interruption resume.

7. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 6, characterized in that, The method of dynamically adjusting hardware test signal parameters using deep reinforcement learning algorithms to compensate for system errors is as follows: Ideal response dataset generation: Based on the distribution network operation standards and twin topology parameters, the virtual terminal generates ideal voltage waveforms, current waveforms, communication message timings, and control command response time datasets for each test scenario. The data sampling interval is consistent with the real terminal detection data. Error calculation: Align the real-time detection data of the actual terminal with the ideal response dataset by timestamp, and calculate the voltage error: Voltage error = |actual voltage value - ideal voltage value| / ideal voltage value × 100%. The current error is calculated in the same way as the voltage error. Communication message delay error =actual message transmission time - ideal message transmission time. Control command response error =actual response delay - ideal response delay. When any error > 0.1%, parameter adjustment is triggered. Deep reinforcement learning adjustment: With all errors ≤0.1% as the optimization objective, a deep reinforcement learning model is constructed, using the DQN algorithm, with a neural network of ≥5 layers and ≥128 hidden layer nodes. The current error value is used as the state input, and the output is the voltage adjustment, current adjustment and frequency adjustment of the wide-range adaptive signal source. Adjustment and verification: The hardware modifies the test signal parameters according to the output adjustment amount, re-acquires terminal detection data and calculates the error. If the error is ≤0.1%, the calibration is completed; if the error is still >0.1%, the above adjustment process is repeated. The maximum number of calibrations in a single test is ≤10, and the total calibration time is ≤1s.

8. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 7, characterized in that, The analysis of terminal defect types based on the multimodal data fusion diagnostic model is as follows: Data preprocessing: The collected abnormal electrical parameter data, abnormal communication message data, and abnormal control behavior data are standardized, mapped to the 0-1 range, noise data is removed, and feature parameters such as electrical parameter fluctuation amplitude, message loss rate, and command response delay time are extracted. Defect type diagnosis: The multimodal data fusion diagnostic model adopts a graph neural network architecture. The preprocessed multimodal feature data is input into the model, and the model assigns weights to each modality of data through an attention mechanism. The weights for electrical parameters are 0.4, communication messages are 0.3, and control behavior is 0.

3. The model outputs the terminal defect type, including hardware failure, software vulnerability, and protocol incompatibility. Fault Root Cause Location: For the diagnosed defect type, the model's built-in defect-root cause association database is retrieved. The defect-root cause association database contains 10,000+ terminal fault cases, covering common fault root causes of terminals of different brands and models. Combined with the hardware model, software version and historical operation and maintenance records of the terminal under test, the specific fault root cause is located. At the same time, the confidence level of the root cause and the assessment of the fault impact range are output.

9. The automated detection platform for a 10kV distribution network intelligent terminal according to claim 8, characterized in that, The specific output process for assessing the root cause confidence level and the scope of fault impact is as follows: Base confidence level acquisition: Retrieve historical fault cases that match the current defect type, terminal hardware model, and software version from the defect-root cause association database, and count the frequency of occurrence of the same fault root cause in the matching cases. Base confidence level = (frequency of occurrence of the same fault root cause / total number of matching cases) × 100%; Feature fit Correction: Compare the electrical parameter fluctuation range, message loss rate and command response delay time corresponding to the abnormal data characteristics of the current terminal with the typical data characteristics corresponding to the root cause of the fault in the matching case, calculate the feature fit, and the corrected confidence = base confidence × (0.6 + 0.4 × feature fit). Historical maintenance weight adjustment: If the terminal under test has historical maintenance records, when the same defect occurred in the historical maintenance and the root cause of the repair is consistent with the current root cause, the confidence level is increased by 5%-10%; if the number of successful repairs in the past is ≥3, the confidence level is increased by 10%; if the number of successful repairs in the past is 1-2, the confidence level is increased by 5%. If the same defect in the historical maintenance corresponds to different root causes, the confidence level is decreased by 3%-8%; if there are ≥2 records of different root causes, the confidence level is decreased by 8%; if there is 1 record of different root causes, the confidence level is decreased by 3%. Final confidence score output: The final confidence score is rounded to one decimal place. The report will also indicate the basis for the confidence score calculation, including the number of matching cases, the feature fit value, and historical operation and maintenance adjustment instructions.

10. An automated detection platform for a 10kV distribution network intelligent terminal according to claim 9, characterized in that, The specific process for generating the fault location analysis based on the multimodal data fusion diagnostic model is as follows: Terminal Function Impact Assessment: Based on defect type and root cause, analyze the impact of each terminal core function, and adopt a three-level judgment standard of normal, degraded and failed. Impact assessment of distribution network operation: Based on the distribution network topology and terminal installation location, assess the impact of defects on distribution network operation indicators, including power supply reliability, power quality and load distribution; Safety risk impact assessment: Identify potential safety hazards caused by defects, including equipment safety, personnel safety and power grid safety, and classify them into three levels: high, medium and low risk. Assessment results output: The assessment results are presented in a structured table format, which includes four columns: impact dimension, specific impact description, impact degree and risk level, and estimated impact scope and consequences. Visual diagrams are also included.

Images