A power distribution network power supply guarantee system

Abstract

This application provides a power distribution network supply assurance system, comprising a multi-source data acquisition unit that collects various types of data from the power distribution network through a standardized interface protocol; an edge-cloud collaborative processing unit that performs real-time data preprocessing, anomaly detection, and power supply prediction generation via edge computing nodes and a cloud analysis server; an intelligent control execution unit that dynamically generates work orders and automatically executes emergency repair operations based on the output; and an adaptive learning and decision-making unit that optimizes system parameters and updates the model. This application achieves the technical effect of comprehensively and efficiently collecting and processing power distribution network data, accurately generating power supply-related results, promptly performing emergency repair operations, optimizing system parameters and models, effectively responding to various faults, and ensuring stable power supply to the power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power supply security technology, and in particular to a power distribution network power supply security system. Background Technology

[0002] In the development of modern society, electricity, as a core element of energy supply, plays a vital role in the normal operation of various industries and people's daily lives. The distribution network, as a key link in power transmission, directly affects the quality of power supply due to its reliability and stability. With continuous economic growth and rapid technological advancements, the number of various electrical devices is constantly increasing, placing higher demands on the intelligent management and power supply guarantee of the distribution network. An efficient and reliable distribution network power supply guarantee system can effectively reduce power outage time and lower the probability of power accidents, thereby ensuring the continuity of industrial production and the convenience of residents' lives. Meanwhile, with the widespread application of information technology, how to fully utilize data to optimize the operation and management of the distribution network has also become a key research focus in the power field. Through data analysis, potential faults can be detected in advance, and timely repair measures can be taken to improve the safety and reliability of the distribution network. Furthermore, an intelligent power supply guarantee system can also achieve the rational allocation of power resources, improve energy utilization efficiency, and promote sustainable development.

[0003] In previous power distribution network protection technologies, to address data acquisition and processing issues, some systems employed diverse acquisition devices, such as sensors and IoT devices, to obtain operational data from the distribution network. Other systems adopted a modular design approach, breaking down the entire system into multiple independent components to improve scalability and maintainability. Still others introduced AI decision-making methods, utilizing artificial intelligence algorithms to analyze and judge the operational status of the distribution network, thereby achieving intelligent management. Regarding data transmission, some systems relied on specific wireless networks, such as LoRa, to enable communication between components. In terms of fault handling, some systems used preset rules and algorithms to generate work orders and execute emergency repair operations.

[0004] However, existing power distribution network protection technologies have significant shortcomings. Some systems lack standardized interfaces, leading to difficulties in data integration and sharing, resulting in data silos and latency issues. Furthermore, the dispersed data acquisition units lack an effective unified coordination mechanism. Some modular systems exhibit poor inter-component coordination, and their communication methods are susceptible to external interference, potentially leading to control conflicts in practical applications. Systems employing AI decision-making often lack interpretability in their decision-making processes, and their state recognition processes are complex and lack real-time performance, making it difficult to respond promptly and accurately to sudden faults. Summary of the Invention

[0005] This invention solves the problems of complex status identification process and low real-time performance in the prior art, and provides a power distribution network power supply guarantee system.

[0006] To achieve the above objectives, the following technical solution is proposed: A power distribution network power supply protection system, comprising: The multi-source data acquisition unit collects real-time operation data, environmental monitoring data, and equipment status data of the power distribution network through standardized interface protocols; The edge-cloud collaborative processing unit includes edge computing nodes and cloud analysis servers. The edge computing nodes are used for real-time data preprocessing and anomaly detection, while the cloud analysis servers are used to generate power supply prediction, equipment health status and fault assessment results. The intelligent control execution unit dynamically generates work orders and automatically executes emergency repair operations based on the output of the edge-cloud collaborative processing unit; And an adaptive learning and decision unit, used to optimize system parameters and update the model based on real-time feedback.

[0007] By adopting the above technical solutions, the standardization and real-time performance of data acquisition are improved, resolving compatibility issues with heterogeneous devices. Standardized interfaces enable plug-and-play functionality and millisecond-level data acquisition. Edge computing nodes perform local preprocessing, reducing cloud load, improving acquisition efficiency, and enhancing compatibility, avoiding the data integration challenges of the original system. Intelligent adaptive control and decision-making are achieved, improving automation and interpretability. Adaptive control algorithms and interpretable AI mechanisms make control decisions more transparent and adaptive, reducing fault repair work order generation time and providing visualized decision reports, thus increasing the trust of maintenance personnel. Overall system integration and cross-technology fusion achieve comprehensive intelligent assurance, deeply integrating IoT architecture, modular design, and AI analysis. Through collaboration, a unified, adaptive, and efficient platform is formed, significantly improving real-time performance, reliability, and intelligence, while reducing maintenance costs.

[0008] Preferably, the multi-source data acquisition unit includes smart sensors, secondary devices, and IoT devices. The multi-source data acquisition unit supports plug-and-play heterogeneous devices and can be configured with an adjustable data acquisition frequency.

[0009] By adopting the above technical solutions, the multi-source data acquisition unit uses a standardized interface protocol to collect data, including smart sensors, secondary devices, and IoT devices. It supports plug-and-play functionality for heterogeneous devices, solving the compatibility problem of heterogeneous devices and enabling plug-and-play functionality. With an adjustable data acquisition frequency, it can achieve millisecond-level data acquisition, improving the standardization and real-time performance of data acquisition. Furthermore, edge computing nodes perform local preprocessing, which can reduce cloud load, enhance compatibility, and avoid the data integration problems of the original system.

[0010] Preferably, the edge computing nodes are deployed locally on the power distribution equipment. The edge computing nodes use stream processing technology to extract and compress electrical transient features in real time, and form a redundant communication link with the cloud analysis server through a 5G link.

[0011] By adopting the above technical solutions, edge computing nodes are deployed locally on power distribution equipment for local preprocessing, which can reduce cloud load and improve data collection efficiency; real-time extraction and compression of electrical transient features are achieved through stream processing technology, enabling rapid data processing; and redundant communication links are formed with the cloud analysis server through 5G links, which enhances communication reliability and avoids malfunctions caused by the loss of wireless signals in the original system.

[0012] Preferably, the edge computing node uses model pruning and knowledge distillation techniques to optimize the AI ​​model.

[0013] By adopting the above technical solutions, edge computing nodes optimize AI models using model pruning and knowledge distillation techniques, enabling AI models to run efficiently on edge devices, reducing resource consumption, lowering training data requirements, improving generalization accuracy, and reducing computational latency. This approach is suitable for millisecond-level response times in high-voltage DC scenarios. Furthermore, combining multi-source data acquisition units that collect data via standardized interface protocols, support plug-and-play functionality for heterogeneous devices, and configure adjustable data acquisition frequencies, along with edge computing nodes deployed locally on power distribution equipment, using stream processing technology to achieve real-time extraction and compression of electrical transient features and establishing redundant communication links with cloud analysis servers via 5G, enhances the standardization and real-time performance of data acquisition, resolves compatibility issues with heterogeneous devices, reduces cloud load, and improves acquisition efficiency and compatibility.

[0014] Preferably, the intelligent control execution unit integrates reinforcement learning algorithms, supports seamless switching between manual and automatic modes, and the switching process is constrained by safety interlock logic.

[0015] By adopting the above technical solutions, the intelligent control execution unit integrates reinforcement learning algorithms, which can dynamically adjust the control strategy according to the working conditions, support seamless switching between manual and automatic modes and be constrained by safety interlocking logic, enhance system coordination and communication reliability, optimize bypass operation and control synchronization, shorten the bypass operation response time to the second level, improve communication reliability, avoid malfunctions caused by the loss of wireless signals in the original system, and improve the system automation level.

[0016] Preferably, the adaptive learning and decision-making unit includes a digital twin module, which is used to simulate the operation scenario of the power distribution network and generate fault contingency plans. The digital twin module outputs the reasoning basis and decision-making scheme through a visual interface.

[0017] By adopting the above technical solutions, combining data acquisition by multi-source data acquisition units, data processing by edge-cloud collaborative processing units, work order generation and emergency repair operations by intelligent control execution units, and system parameter optimization and model updating by adaptive learning and decision-making units, the digital twin module of the adaptive learning and decision-making unit simulates power distribution network operation scenarios and generates fault contingency plans. It can formulate solutions to deal with faults in advance, and output the reasoning basis and decision-making scheme through a visual interface, making control decisions more transparent and explainable, and improving the trust of operation and maintenance personnel.

[0018] Preferably, it also includes a cable switching and monitoring module, which includes an automatically aligned smart connector and an AR-assisted installation component, the AR-assisted installation component providing three-dimensional positioning guidance via a handheld terminal.

[0019] By adopting the above technical solutions, the multi-source data acquisition unit collects data through standardized interface protocols, improving the standardization and real-time performance of data acquisition and solving the compatibility problem of heterogeneous devices. On this basis, the multi-source data acquisition unit includes intelligent sensors and supports plug-and-play functionality for heterogeneous devices and adjustable data acquisition frequencies, further enhancing compatibility and acquisition flexibility. The cable transfer and monitoring module includes an automatically aligned intelligent connector and an AR-assisted installation component that provides 3D positioning guidance via a handheld terminal, simplifying the cable transfer and installation process, reducing reliance on manual labor and safety risks, shortening cable transfer time, and reducing installation error rates.

[0020] Preferably, the cable switching and monitoring module has a built-in self-diagnostic unit that continuously monitors the connector contact impedance and triggers an alarm mechanism for abnormal conditions.

[0021] By adopting the above technical solutions, the system's cable transfer and monitoring module has a built-in self-diagnostic unit that can continuously monitor the contact impedance of the joints and trigger an abnormal state alarm mechanism. It can monitor the joint status in real time, and combined with safety interlock logic, it avoids the risk of misoperation in the original system's manual mode. At the same time, the entire power distribution network power supply guarantee system collects data through a multi-source data acquisition unit, performs data processing and analysis through an edge-cloud collaborative processing unit, generates work orders and executes emergency repair operations through an intelligent control execution unit, and optimizes system parameters and updates the model through an adaptive learning and decision-making unit. This improves the standardization and real-time performance of data acquisition, the system's coordination and communication reliability, and the intelligent adaptive control and decision-making capabilities. It simplifies the cable transfer and installation process, optimizes computational efficiency and model adaptability, and achieves overall system integration and cross-technology fusion. It has significantly improved in terms of real-time performance, reliability, and intelligence, while reducing operation and maintenance costs.

[0022] Preferably, the edge-cloud collaborative processing unit achieves instruction synchronization between units through a distributed consensus algorithm and uses an event-driven architecture to trigger the entire process analysis.

[0023] By adopting the above technical solutions, the system's coordination and communication reliability are enhanced, bypass operation and control synchronization are optimized, and the synchronization of instructions of each unit is ensured. Combined with reinforcement learning control, the bypass operation response time can be shortened to the second level, communication reliability is improved, and the original system's malfunctions caused by the loss of wireless signals are avoided. Furthermore, the event-driven architecture triggers the full-process analysis only when a significant transient is detected.

[0024] Preferably, the adaptive learning and decision-making unit integrates historical operating condition samples and real-time data, and improves the ability to respond to and adapt to new failure modes through transfer learning.

[0025] By adopting the above technical solutions, historical working condition samples and real-time data are integrated, and transfer learning is used to improve the system's ability to respond to and adapt to new failure modes, optimize computational efficiency and model adaptability, enable the AI ​​model to quickly adapt to new scenarios, reduce training data requirements, and improve generalization accuracy.

[0026] The beneficial effects of this invention are: 1. Improve the standardization and real-time performance of data acquisition, solve the compatibility problem of heterogeneous devices, collect data through standardized interface protocols, support plug-and-play of heterogeneous devices and millisecond-level data acquisition, perform local preprocessing on edge computing nodes, reduce cloud load, improve acquisition efficiency, enhance compatibility, and avoid the data integration problems of the original system.

[0027] 2. Enhance system coordination and communication reliability, optimize bypass operation and control synchronization, realize instruction synchronization between units through distributed consensus algorithm, and adopt event-driven architecture to trigger the whole process analysis process. Combined with reinforcement learning control, the bypass operation response time is shortened to the second level, communication reliability is improved, and the malfunction caused by the loss of wireless signal in the original system is avoided.

[0028] 3. Achieve intelligent adaptive control and decision-making, improve automation level and interpretability. Through adaptive control algorithms and interpretable AI mechanisms, control decisions become more transparent and adaptive, the time for generating fault repair work orders is reduced, and the decision report is visualized, which enhances the trust of operation and maintenance personnel. Through model optimization, computational latency is reduced, making it more suitable for the high real-time requirements of flexible DC systems.

[0029] 4. Simplify cable switching and installation processes, reduce reliance on manual labor and safety risks. Through automatic alignment smart connectors and AR-assisted installation components, cable switching time is shortened and installation error rate is reduced. The self-diagnostic unit continuously monitors the connector contact impedance and triggers an abnormal status alarm mechanism. Combined with safety interlock logic, it avoids the risk of misoperation in the original system's manual mode.

[0030] 5. Optimize computational efficiency and model adaptability, improve processing speed and generalization ability. Through model pruning and transfer learning, enable AI models to run efficiently on edge devices, reduce resource consumption, and quickly adapt to new scenarios such as extreme weather. Reduce training data requirements and improve generalization accuracy. Attached Figure Description

[0031] Figure 1 This is a diagram showing the connection relationships between the system modules of the present invention.

[0032] Figure 2 for Figure 1 Diagram showing the connection relationships between sub-units in a multi-source data acquisition unit.

[0033] Figure 3 for Figure 1 Diagram showing the connection relationships between sub-units in the edge-cloud collaborative processing unit.

[0034] Figure 4 for Figure 1 Connection diagram of sub-units in the cable switching and monitoring module. Detailed Implementation

[0035] The technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments are only possible technical implementations of the present invention, but are not limited thereto. Other embodiments obtained by those skilled in the art in conjunction with the embodiments of the present invention without creative effort are also within the protection scope of the present invention. Example

[0036] This application provides a power distribution network supply assurance system, including a multi-source data acquisition unit, an edge-cloud collaborative processing unit, an intelligent control execution unit, and an adaptive learning and decision-making unit. These units cooperate to improve the compatibility, real-time performance, coordination, and interpretability of the power distribution network supply assurance system. This is because the multi-source data acquisition unit collects real-time operational data, environmental monitoring data, and equipment status data of the power distribution network through standardized interface protocols, solving the compatibility problem of heterogeneous equipment. The edge-cloud collaborative processing unit divides tasks, reducing cloud load and enabling multi-dimensional analysis. The intelligent control execution unit can dynamically generate work orders and automatically execute emergency repair operations based on the output of the edge-cloud collaborative processing unit. The adaptive learning and decision-making unit can optimize system parameters and update the model based on real-time feedback.

[0037] The multi-source data acquisition unit includes smart sensors, secondary devices, and IoT devices, supporting plug-and-play functionality for heterogeneous devices and configurable adjustable data acquisition frequencies.

[0038] Intelligent sensors employ high-precision current and voltage sensors, typically characterized by small size and high accuracy, enabling them to accurately collect real-time operational data from power distribution networks, such as the magnitude and changes in current and voltage. In addition to common electromagnetic sensors, this application can also utilize fiber optic sensors, which offer the advantage of strong anti-interference capabilities.

[0039] Secondary equipment includes protection devices, measurement and control devices, etc., and is mainly used to protect and monitor equipment in the power distribution network. Protection devices can act promptly when equipment malfunctions, cutting off the circuit and preventing the fault from escalating. The secondary equipment in this application can also be digital intelligent secondary equipment, characterized by strong communication capabilities and fast processing speed.

[0040] IoT devices are terminal devices with communication capabilities, such as smart meters, that can upload device status data in real time. IoT devices can also utilize IoT gateways with wireless communication capabilities, which can connect multiple devices to achieve centralized data collection and transmission.

[0041] When the system starts up, it broadcasts a discovery message based on IEC 61850 or MQTT.

[0042] The newly connected device responds with a message and reports its device description file, which contains data point identifiers, such as MMXU1.PhV.phsA.cVal.mag.f, data type, acquisition accuracy, and supported minimum / maximum acquisition frequencies F_min and F_max.

[0043] The system dynamically sets the acquisition frequency F_set for each data point based on scenario strategies, such as steady-state monitoring, transient capture, and fault recording. The logical judgment is as follows: IF Scenario == "Steady-state monitoring" THEN F_set = F_min(eg, 1Hz) ELSE IF Scene == "Transient Capture" THEN F_set=(F_min+F_max) / 2(eg,1kHz) ELSE IF Event triggered (e.g., overcurrent alarm) THEN F_set=F_max(eg,10kHz) / / Enter high-frequency recording mode END IF The collected raw data, such as voltage U(t) and current I(t), is encapsulated into standardized data frames: Frame={Timestamp,Device_ID,Data_Point_ID,Data_raw,Data_Quality} It is published to the data bus through a unified communication interface for edge computing nodes to subscribe to.

[0044] The multi-source data acquisition unit integrates and transmits real-time operational data, environmental monitoring data, and equipment status data of the power distribution network through a standardized interface protocol, providing an accurate data foundation for subsequent processing and analysis and solving the compatibility problem of heterogeneous equipment. By dynamically adjusting F_set, bandwidth resources are optimized while ensuring data real-time performance, unlike existing technologies that rely on fixed-frequency acquisition.

[0045] The edge-cloud collaborative processing unit includes edge computing nodes and cloud analytics servers.

[0046] Edge computing nodes are deployed locally on power distribution equipment. They employ embedded computers and are characterized by their small size and low power consumption. Edge computing nodes achieve real-time extraction and compression of electrical transient features through stream processing technology. For example, by using high-speed data processing chips and corresponding algorithms, electrical transient features can be quickly extracted from large amounts of data and compressed to reduce data transmission volume.

[0047] Input a high-frequency sampling sequence X={x1,x2,...,xn}, such as voltage values.

[0048] The feature vector V_edge is calculated using a streaming processing algorithm.

[0049] For example, calculating the effective value, harmonic distortion rate, and drop depth ΔU within a time window: Fall depth ΔU=(U_nominal-U_min) / U_nominal In the formula, U_min is the minimum voltage within the window, and U_nominal is the rated voltage.

[0050] Output the compressed feature vector V_edge and the anomaly flag Flag_abnormal. If ΔU > threshold, set the flag.

[0051] Edge computing nodes also employ model pruning and knowledge distillation techniques to optimize AI models.

[0052] The compact convolutional neural network (CNN) obtained by knowledge distillation is used as input V_edge, and outputs the preliminary fault classification probability P_edge={p_edge_fault1,p_edge_fault2,...} and the preliminary equipment health score H_edge.

[0053] In addition to embedded computers, the edge computing nodes in this application can also use industrial-grade single-board computers, which offer more stable and reliable performance.

[0054] Edge computing nodes and cloud analytics servers form redundant communication links via 5G links. In the event of a 5G network failure, satellite links can be used to ensure reliable data transmission. The cloud analytics server employs a high-performance server cluster with powerful computing and storage capabilities. It performs deep learning and multi-dimensional analysis based on historical data to generate power supply predictions, equipment health status, and fault assessment results.

[0055] The multi-dimensional deep analysis model of the cloud analytics server includes: The power supply prediction model is based on the spatiotemporal graph neural network (ST-GNN). It takes the whole network topology, load history, weather data W, and edge-uploaded data V_edge as input, and outputs the power outage probability P_blackout(node_i,T) of each node in the future time T. The equipment health assessment model integrates historical operation and maintenance records M_history and real-time operating conditions V_edge, and calculates the remaining useful life (RUL) and health index H_cloud of the equipment through a survival analysis model. The accurate fault assessment model employs multimodal deep learning, integrating electrical features V_edge, protection device action signals S_protection, and environmental data E to output a comprehensive assessment result Result_fault, which includes the cause, location, and severity of the fault.

[0056] Edge computing nodes perform real-time data preprocessing and anomaly detection, and then transmit the processed data to a cloud analytics server. The cloud analytics server then performs further analysis and processing. The two work together to improve the efficiency and accuracy of data processing.

[0057] The edge-cloud collaborative processing unit also achieves instruction synchronization between units through a distributed consensus algorithm and uses an event-driven architecture to trigger the entire process analysis.

[0058] Distributed consensus algorithms aim to solve the problems of instruction synchronization and coordination. They are applied when multiple edge nodes or control units need to reach a consensus on the same control instruction, thus avoiding single points of failure or instruction conflicts.

[0059] Specific logic: Role: One leader, multiple followers.

[0060] Synchronization process: The leader receives the cloud command Cmd and appends it as a log entry to the local machine. The leader sends an AppendEntries RPC, containing the instruction Cmd, to all followers in parallel. Once received, the follower verifies the validity of the Cmd, such as its signature and timing. If valid, it appends log entries and replies with a success message. Once the leader receives successful responses from more than half of the followers, they submit the log, indicating that the instruction has reached a consensus, and then notify the followers to submit it. All nodes execute the Cmd command according to the committed log.

[0061] Distributed consensus algorithms ensure strong consistency and high availability of control commands in a multi-node environment. Even if individual nodes fail, the system can still reach a consensus and coordinate actions based on the majority, fundamentally solving the problem of unreliable coordination in distributed systems.

[0062] Event-driven architecture aims to trigger full-process analysis and improve real-time performance. It is applied to defining all state changes in the system as events, such as Event_VoltageDip(ΔU,location,time) and Event_TempHigh(T,device_ID).

[0063] Event bus: Establish a unified event publish / subscribe bus.

[0064] Triggering logic: The data acquisition unit detects an anomaly and publishes Event_X to the event bus; Edge computing nodes subscribe to relevant events, which trigger their feature extraction and lightweight inference processes, and publish a new event Event_EdgeResult(V_edge,Flag); The cloud analytics server subscribes to Event_EdgeResult, and the event triggers its multi-dimensional in-depth analysis process. The intelligent control unit subscribes to Event_Cmd(Cmd) in the cloud, and the event triggers its control execution process.

[0065] Event-driven architecture enables a loosely coupled, highly cohesive system design. Modules respond asynchronously and concurrently, and processes are automatically linked by events, eliminating polling latency and transforming system response from "periodic inspection" to "event triggering," greatly improving real-time performance.

[0066] The intelligent control execution unit integrates reinforcement learning algorithms, supporting seamless switching between manual and automatic modes. Employing devices such as programmable logic controllers, it can dynamically generate work orders and automatically execute emergency repair operations based on the output of the edge-cloud collaborative processing unit. The reinforcement learning algorithm enables the intelligent control execution unit to dynamically adjust its control strategy according to different operating conditions, improving control flexibility and adaptability.

[0067] Reinforcement Learning (RL) Algorithm Decision Logic: State space (S): s_t=[Grid_State_t,Event_Type,Device_Status,Mode_current], which includes the real-time state of the power grid, event type, device status, and current mode.

[0068] Action space (A): a_t includes {generating work orders Dispatch_WorkOrder(X), executing switch opening and closing Operate_Switch(Y), switching running mode Switch_Mode(Z), ...}.

[0069] Reward function (R) design: R(t) = w1 * (-Δt_restoration) / / Reward for rapid recovery +w2*(-Cost_operational) / / Rewards for reducing operational costs +w3*(Reliability_improvement) / / Reward reliability improvement +Penalty_if_unsafe / / Severe penalties for security violations Based on the current state s_t, the agent outputs the optimal action a_t through the trained policy network π(a|s). For example, in the state of "voltage drop + overload", the policy network may output the action "Operate_Switch (switch to backup line L_backup)".

[0070] The switching process between manual and automatic modes is constrained by safety interlock logic to ensure the safety of the switching process. For example, in manual mode, the operator can operate according to the actual situation; in automatic mode, the intelligent control execution unit can automatically perform the corresponding operations. When it is necessary to switch from manual mode to automatic mode, the safety interlock logic checks whether the current state meets the switching conditions. If it does, the switch is performed; otherwise, a prompt is given.

[0071] Interlocking condition set: Defines a set of conditions, Conditions_auto, that must all be met before switching to automatic mode is allowed.

[0072] Conditions_auto={ C1: All critical devices are communicating normally = TRUE C2: Model confidence > threshold confidence. C3: No human is currently performing any operations = TRUE C4: The system is in a stable state (no drastic fluctuations) = TRUE } Seamless switching logic flow: Mode switching request: Mode switching is triggered by operator request or RL decision.

[0073] Security check: The system calculates Conditions_auto in real time. If AND(Conditions_auto) == TRUE, proceed to step 3; otherwise, lock the switch and highlight the specific conditions that are not met on the human-machine interface (HMI).

[0074] Smooth transfer of control: Manual → Automatic: The system first synchronizes the setpoints of all current manual control commands and the device state snapshot to the RL agent as its initial state s_0. Subsequently, control is transferred, and the RL outputs action a_t.

[0075] Automatic → Manual: When the operator operates on the HMI, the system immediately freezes the RL output, and control is instantly returned to the operator. Simultaneously, the RL agent begins recording and learning the sequence of manual operations as new training samples, thus learning from human experience.

[0076] Reinforcement learning provides dynamic optimization capabilities, enabling control strategies to adapt to complex and ever-changing power grid conditions. Safety interlocking logic provides rigid safety guarantees, preventing accidental switching of automatic modes under unsafe conditions. The combination of these two elements achieves a unity of "intelligent flexibility" and "safety rigidity," maximizing automation benefits while ensuring absolute safety, and realizing truly reliable "seamless switching."

[0077] The adaptive learning and decision-making unit includes a digital twin module. This module combines virtual simulation software and hardware to simulate power distribution network operation scenarios and generate contingency plans. It establishes a virtual model of the power distribution network, reflecting its actual operating status in real time.

[0078] The adaptive learning and decision-making unit also integrates historical operating condition samples and real-time data to improve its ability to adapt to new failure modes through transfer learning.

[0079] The digital twin module outputs the reasoning basis and decision-making solutions through a visual interface, making it convenient for operation and maintenance personnel to view and analyze them. For example, operation and maintenance personnel can intuitively understand the operation status of the distribution network and the specific content of the fault contingency plan through the visual interface.

[0080] Real-time simulation logic of digital twin module: State synchronization: The initial state of the digital twin virtual model M_virtual is synchronized with the physical entity M_physical in real time through protocols such as OPC UA: State_virtual(t0) = State_physical(t0).

[0081] Fault simulation and contingency plan generation: When an Event_Fault is received, the same fault is injected into M_virtual.

[0082] Run real-time simulation calculations to solve the differential algebraic equations of the power grid and predict fault evolution paths.

[0083] Based on a predefined optimization objective, multiple control schemes (Plan_i) are tried in parallel during simulation, and their effects (Effect(Plan_i)) are evaluated.

[0084] Output the optimal contingency plan and its visual reasoning chain: Recommendation=ArgMax(Effect(Plan _i)).

[0085] The adaptive learning and decision-making unit can also integrate historical operating condition samples and real-time data to improve its ability to adapt to new failure modes through transfer learning. It can update the model based on real-time feedback, optimize system parameters, and make the system more intelligent and adaptive.

[0086] Transfer learning model update logic: Process: When a new fault F_new occurs, the system collects the handling data D_new = {s, a, r,s'}.

[0087] Update: Model-based transfer learning is used, where the pre-trained model M_pre is fine-tuned on D_new with a few iterations, instead of being trained from scratch. The loss function includes a term for fitting the new data and a term for retaining old knowledge. Loss_total=α*Loss_new(M,D_new)+β*Loss_old(M,M_pre) Effect: Quickly adapts to new faults while avoiding "catastrophic amnesia".

[0088] The system of this application also includes a cable splicing and monitoring module, which comprises an automatically aligned smart connector and an AR-assisted installation component. The automatically aligned smart connector has an automatic alignment function, enabling fast and accurate cable connection, reducing manual operation time and errors. It employs a high-precision mechanical structure and sensors to ensure connector alignment accuracy.

[0089] Automatic alignment smart connector control logic: Pose awareness: The connector has a built-in vision sensor / LiDAR to acquire point cloud data P_cloud of the terminal to be connected.

[0090] Pose calculation: The transformation matrix T between the current joint pose Pose_current and the target pose Pose_target is calculated using algorithms such as ICP (Iterative Closest Point).

[0091] Motion control: Control the micro servo motor to make the joint move along the translation and rotation amounts decomposed by the T matrix until the pose error ‖Pose_current-Pose_target‖ < ε, where ε is the docking accuracy threshold.

[0092] Connection confirmation: The physical connection is confirmed by the impedance / pressure value fed back by the contact sensor.

[0093] The AR-assisted installation component provides 3D positioning guidance via a handheld terminal. Operators can visually see the cable's installation location and orientation through the AR interface on the handheld terminal, improving installation accuracy. For example, when an operator needs to install a cable, the AR-assisted installation component can display the correct installation position and angle of the cable on the handheld terminal's screen, and the operator only needs to follow the guidance.

[0094] AR-assisted installation component logic: Spatial registration: Using SLAM technology, the coordinate system of the handheld terminal is aligned with the coordinate system of the real world.

[0095] 3D Guided Generation: The digital twin provides the standard installation pose (Pose_standard) for the cable connector. The AR component calculates the overlay relationship between Pose_standard and the operator's current field of view, rendering virtual cable outlines, docking arrows, torque values, and other guiding information, which are then overlaid on the real-world scene in real time.

[0096] The cable connection and monitoring module has a built-in self-diagnostic unit that continuously monitors the connector contact impedance and triggers an alarm mechanism for abnormal conditions.

[0097] Self-diagnostic unit alarm logic: Continuous monitoring: Real-time measurement of the connector contact impedance Z_contact(t).

[0098] Health model: Establish the impedance baseline Z_baseline and the allowable fluctuation range ΔZ(t) as a function of time and load.

[0099] Alarm triggered: IF Z_contact(t)>Z_baseline+ΔZ(t)+Margin THEN Status=“Early Warning” IF The trend continues to deteriorate (dZ / dt>threshold) THEN Status="Alert", triggering Event_ConnectionAlert END IF END IF The self-diagnostic unit enables precise, rapid, and monitorable cable connections, transforming the traditionally experience-dependent work into a standardized, digitalized, and predictive automated process, greatly reducing safety risks and operation time.

[0100] The implementation principle of this embodiment is as follows: This power distribution network power supply guarantee system effectively solves the defects of existing technologies in terms of compatibility, real-time performance, collaboration, and interpretability through the collaborative work of various units. The multi-source data acquisition unit adopts a standardized interface protocol, realizing plug-and-play functionality for heterogeneous devices and improving the compatibility and real-time performance of data acquisition; the edge-cloud collaborative processing unit reduces cloud load and improves data processing efficiency through the division of labor between edge computing nodes and cloud analysis servers; the intelligent control execution unit integrates reinforcement learning algorithms, realizing seamless switching between dynamic control and manual-automatic modes, improving control flexibility and security; the adaptive learning and decision-making unit enhances the system's adaptability to new fault modes and the interpretability of decisions through digital twin modules and transfer learning. The overall system forms a unified, adaptive, and efficient platform, which has significant improvements in real-time performance, reliability, and intelligence compared to existing technologies, while reducing operation and maintenance costs. The cable transfer and monitoring module simplifies the cable transfer and installation process, reducing reliance on manual labor and safety risks. The automatic alignment function of the automatic alignment intelligent connector and the three-dimensional positioning guidance of the AR-assisted installation component shorten cable transfer time and reduce installation error rate. Meanwhile, the cable transfer and monitoring module has a built-in self-diagnostic unit that continuously monitors the contact impedance of the joint and triggers an alarm mechanism for abnormal conditions. This enables timely detection of joint abnormalities, prevents faults from occurring, and further improves the reliability and safety of the power distribution network. Example

[0101] Taking a pumped storage power station in the East China Power Grid as an example, a lightning strike caused a momentary ground fault on the ±500kV DC bus, resulting in a sudden drop in system voltage. This embodiment will demonstrate in detail the entire process, specific calculations, and decision-making logic of a power distribution network supply guarantee system provided in this application in handling this event.

[0102] The system has a known initial state: Rated DC voltage U_nominal = 500kV Voltage sag alarm threshold U_threshold = 485kV The communication status of critical equipment is normal.

[0103] The system operates in the "steady-state monitoring" mode, and the data acquisition frequency F_set = 1 Hz.

[0104] Step S1: Event triggering and high-speed data acquisition (0 - 50 milliseconds) Event occurrence: A lightning strike causes the bus voltage to drop from 500 kV to 480 kV within 5 milliseconds.

[0105] Response of the data acquisition unit: Frequency dynamic adjustment: The intelligent voltage sensor deployed on the bus side detects that U(t) = 480 kV < U_threshold, immediately generates the Event_VoltageDip(ΔU = 20 kV, Location = BusA, Time = 14:30:00.000) event locally, and publishes it through the event bus. Meanwhile, according to the preset logic: IF event is triggered (overvoltage / undervoltage alarm) THEN F_set = F_max = 10 kHz / / Switch to the "fault recording" mode END IF High-speed acquisition and encapsulation: The sensor acquires the instantaneous voltage value sequence U_raw(t) at a frequency of 10 kHz and encapsulates it into a standardized frame: Frame={Timestamp:14:30:00.005,Device_ID:VSensor_BusA_01,Data_Point_ID:MMXU1.DCV.phsA.cVal.mag.f,Data_raw:480.2,Data_Quality:Good} Step S2: Real-time feature extraction by the edge computing node (50 - 150 milliseconds) Subscription and processing: The edge computing node subscribes to Event_VoltageDip and immediately starts stream processing. [[ID=二十三]]

[0106] Feature vector calculation: Analyze the voltage sampling sequence X={x1,x2,...,x200} (sampled at 10 kHz) of the most recent cycle (20 ms).

[0107] Calculate the effective value: U_rms = sqrt(mean(X^2)) ≈ 480.5 kV Calculate the drop depth: ΔU=(U_nominal - U_min) / U_nominal=(500 - 480.0) / 500 = 4.0% Calculate the harmonic distortion rate (taking the 5th harmonic as an example): Through FFT calculation, THD5 ≈ 2.1% (significantly increased compared to 0.5% in the steady state).

[0108] Feature vector: V_edge={U_rms:480.5,ΔU: 4.0%, THD5: 2.1%, fault initiation phase: 89°} Lightweight AI Inference: Input V_edge into a lightweight CNN model optimized by knowledge distillation. The model outputs preliminary probability assessment: P_edge = {Instantaneous grounding: 85%, Commutation failure: 10%, Other: 5%}, and health score H_edge = 65 (decreasing). Then, a new event is published: Event_EdgeResult(V_edge, Flag_abnormal=TRUE, P_edge).

[0109] Step S3: Cloud-based deep analysis and collaborative decision-making (150-800 milliseconds) Multi-model in-depth analysis: Power supply prediction (ST-GNN model): Input: network topology, load data, weather forecast (thunderstorms), V_edge. Model output: the probability of power outage for load center L1 affected by this power station within the next 30 minutes: P_blackout(L1,30min)=42%.

[0110] Equipment health assessment (Cox model): Input the historical breakdown records of this bus (3 times), current V_edge, and ambient humidity (95%). The model calculates a risk ratio HR=3.2, the remaining useful life RUL is revised from 10 years to 8 years, and the health index H_cloud=60.

[0111] Accurate Fault Assessment (Multimodal Model): Integrating V_edge, traveling wave protection device action signal (not activated), and lightning location system data (hit point 300 meters from busbar). The model performs comprehensive calculations and outputs the final assessment: Result_fault = {Fault type: Instantaneous grounding caused by lightning backflashover; Fault location: Busbar A insulator; Confidence level: 92%}.

[0112] Generate control commands: Based on the analysis results, the cloud platform selects the best option from three contingency plans through digital twin simulation. Plan 1: Immediately trip the DC circuit breaker at this station, resulting in a total power loss. Assessment of effect: Significant power outage losses.

[0113] Plan 2: Start the backup diesel generator to provide short-term power support for the substation. Assessment result: A short-term power outage occurred.

[0114] Plan 3: Control the adjacent grid-connected photovoltaic power station PV_Farm to rapidly increase its output by 50MW, and provide dynamic voltage support through the on-site SVC (Static Var Compensator). Digital twin simulation shows that this scheme can stabilize the voltage and has the lowest overall cost.

[0115] Decision: Recommendation = Plan3. Generate control command: Cmd = {Control_Target:PV_Farm&SVC_BusA;Action:Increase_Output_by(50MW)&Boost_Reactive_Power;Valid_Time:Immediate}.

[0116] Distributed consensus achieved: Roles: The cloud server is the command initiator (Leader), and the PV_Farm edge controller and the local SVC controller are the followers (Follower1, Follower2).

[0117] process: The leader appends Cmd as a log entry Log_Index=101 to the local log.

[0118] The Leader sends AppendEntries RPC(Log_Index=101,Cmd) to Follower1 and Follower2 in parallel.

[0119] After verifying the Cmd signature and timing, the two Followers append Log_101 to their respective logs and reply with "Success".

[0120] The Leader receives more than half (2 / 2) of the successful responses and commits Log_101.

[0121] The Leader notifies Follower1 and Follower2 to commit Log_101. At this point, the three parties have reached a distributed consensus on "executing Cmd".

[0122] Step S4: Intelligent control execution and safe switching (800-1500 milliseconds) Event-triggered execution: After Cmd is submitted, it is published to the event bus as Event_Cmd(Cmd).

[0123] Safety interlock verification: The intelligent control unit of this site subscribes to this event. Before execution, verify Conditions_auto: C1: Is communication with PV_Farm and SVC normal? =TRUE.

[0124] C2: Cloud model confidence 92% > threshold (85%)? = TRUE.

[0125] C3: No human is currently operating the system? =TRUE.

[0126] C4: Voltage transient has stabilized (fluctuation <1%)? =TRUE.

[0127] Result: AND(Conditions_auto)=TRUE, allowing automatic execution.

[0128] Strengthening learning decision-making and execution: Status: s_t=[U=485kV (recovering),Event_Type=Instantaneous grounding,SVC_Status=Standby,Mode=Auto] Policy network: Based on s_t, the policy network π(a|s) outputs the action probability, with the highest being a_t=Execute(Cmd).

[0129] Execution: The control unit sends precise control commands to PV_Farm and SVC via a safety protocol. PV_Farm will increase its output by 50MW within 200ms, and SVC will engage capacitive reactive power compensation.

[0130] Feedback: Within 1500 milliseconds, the bus voltage recovered to U(t) = 498kV. The system was safe and stable, avoiding load loss.

[0131] Step S5: Preventive maintenance of the cable switching and monitoring module (after the fact, 30 minutes later) Suppose that the digital twin predicts that the busbar insulator needs to be replaced in a planned manner, triggering a maintenance work order.

[0132] AR-assisted installation: Maintenance personnel wearing AR glasses approach the connector. The SLAM algorithm completes spatial registration and establishes a virtual coordinate system.

[0133] The digital twin sends the standard mounting pose Pose_standard (containing three-dimensional coordinates [X=1000mm, Y=500mm, Z=200mm] and Euler angles [0°, 0°, 30°]) to the AR glasses.

[0134] The AR component calculates and overlays a semi-transparent green virtual connector outline above the real connector in the field of view, with an animated arrow pointing to the docking position and the text prompt "Rotate 30° to tighten".

[0135] Automatic alignment and connection: The LiDAR scanner on the smart connector acquires the real terminal point cloud P_cloud.

[0136] The transformation matrix T between the current connector pose Pose_current=[990, 510, 205, 0°, 5°, 10°] and Pose_standard is obtained through iterative calculation using the ICP algorithm.

[0137] The control system calculates T to obtain the adjustment amounts: ΔX=+10mm, ΔY=-10mm, ΔZ=-5mm, Δθz=+20°, and drives the micro servo motor to complete the alignment.

[0138] The contact sensor reports that the pressure has reached the set value, confirming a reliable connection.

[0139] Self-diagnostic monitoring: After connection, the self-diagnostic unit continuously monitors the contact impedance Z_contact(t). The initial baseline Z_baseline = 5 μΩ, the allowable fluctuation ΔZ(t) = 0.5 μΩ, and the warning margin Margin = 0.2 μΩ.

[0140] In the 5th hour, Z_contact=5.8μΩ>(5+0.5+0.2)=5.7μΩ was detected, and the status changed to "warning".

[0141] The system continuously monitors the trend and finds that dZ / dt=0.15μΩ / h>threshold (0.1μΩ / h), so the status is upgraded to "alarm" and Event_ConnectionAlert is triggered to prompt maintenance personnel to tighten the bolts.

[0142] Step S6: Adaptive Learning (Post-hoc) All data D_new from this failure was stored.

[0143] The adaptive learning unit is initiated, using D_new to fine-tune the fault assessment model through transfer learning. In the loss function, Loss_new ensures the model can better identify the "lightning backlash" feature, while Loss_old prevents the model from forgetting other faults such as "pollution flashover grounding". After a few iterations, the model's confidence in assessing new fault modes is expected to improve by 5%.

[0144] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A power distribution network supply guarantee system, characterized in that, include: The multi-source data acquisition unit collects real-time operation data, environmental monitoring data, and equipment status data of the power distribution network through standardized interface protocols; The edge-cloud collaborative processing unit includes an edge computing node and a cloud analysis server. The edge computing node is used for real-time data preprocessing and anomaly detection, and the cloud analysis server is used to generate power supply prediction, equipment health status and fault assessment results. The intelligent control execution unit dynamically generates work orders and automatically executes emergency repair operations based on the output of the edge-cloud collaborative processing unit. And an adaptive learning and decision unit, used to optimize system parameters and update the model based on real-time feedback.

2. The power supply guarantee system of a power distribution network according to claim 1, characterized in that, The multi-source data acquisition unit includes intelligent sensors, secondary devices, and IoT devices. The multi-source data acquisition unit supports plug-and-play heterogeneous devices and can be configured with an adjustable data acquisition frequency.

3. The power supply guarantee system of an electric power distribution network according to claim 2, characterized in that, The edge computing node is deployed locally on the power distribution equipment. The edge computing node uses stream processing technology to extract and compress electrical transient features in real time, and forms a redundant communication link with the cloud analysis server through a 5G link.

4. The power supply guarantee system of an electric distribution network according to claim 2 or 3, characterized in that, The edge computing nodes employ model pruning and knowledge distillation techniques to optimize the AI ​​model.

5. The power supply guarantee system of an electric power distribution network according to claim 1, characterized by, The intelligent control execution unit integrates reinforcement learning algorithms and supports seamless switching between manual and automatic modes. The switching process is constrained by safety interlocking logic.

6. The power supply guarantee system of an electric distribution network according to claim 1, characterized in that, The adaptive learning and decision-making unit includes a digital twin module, which is used to simulate power distribution network operation scenarios and generate fault contingency plans. The digital twin module outputs the reasoning basis and decision-making scheme through a visual interface.

7. The power supply guarantee system of an electric distribution network according to claim 1, characterized in that, It also includes a cable switching and monitoring module, which includes an automatically aligned smart connector and an AR-assisted installation component, the AR-assisted installation component providing three-dimensional positioning guidance via a handheld terminal.

8. A power distribution network supply guarantee system according to claim 7, characterized in that, The cable switching and monitoring module has a built-in self-diagnostic unit, which continuously monitors the connector contact impedance and triggers an alarm mechanism for abnormal conditions.

9. The power supply guarantee system of an electric distribution network according to claim 1, characterized in that, The edge-cloud collaborative processing unit achieves instruction synchronization between units through a distributed consensus algorithm and uses an event-driven architecture to trigger the entire analysis process.

10. The power supply guarantee system of an electric distribution network according to claim 1, characterized in that, The adaptive learning and decision-making unit integrates historical operating condition samples and real-time data, and improves its ability to respond to and adapt to new failure modes through transfer learning.

Images