Cloud-edge collaborative hybrid simulation test method and platform for feeder automation

Abstract

The application discloses a kind of cloud edge coordination-based feeder automation hybrid simulation test method and platform.Method includes: constructing cloud digital twin main simulation and edge hardware-in-the-loop hybrid model;According to strategy, cloud decision area and edge control area are divided and collaborative task sequence is generated;According to edge real-time capability, simulation step, interface mapping and clock reference are dynamically determined;With fault timestamp sequence, start global clock;Drive hardware-in-the-loop to collect terminal action and electrical quantity, and cover edge time stamp feedback;Time-space alignment with time-stamped data, reference vector interpolation compensation and advance global clock according to slowest edge node;Evaluate the fault location accuracy, isolation coordination and power restoration time of strategy.Platform includes cloud main simulation module, collaborative management module and edge semi-physical simulation module.The application can reproduce cloud edge timing error and resource competition, improve test fidelity.

Description

Technical Field

[0001] This application relates to the field of distribution network automation testing and simulation technology, specifically to a cloud-edge collaborative hybrid simulation testing method and platform for feeder automation. Background Technology

[0002] As the requirements for power supply reliability and self-healing capabilities in power distribution networks continue to increase, feeder automation technology is gradually evolving from the traditional local control mode to a cloud-edge collaborative architecture of "centralized decision-making in the cloud and local control at the edge." In this cloud-edge collaborative model, the cloud generates and distributes fault handling strategies based on global information, while the distribution terminals on the edge nodes perform protection actions based on local real-time data. The two work closely together through a communication network to jointly complete fault location, isolation, and power restoration. However, while this collaborative model brings flexibility and increased intelligence, it also introduces new challenges such as differences in edge node computing capabilities, communication latency fluctuations, and cloud-edge clock asynchrony, placing higher demands on the thorough verification of feeder automation strategies before actual deployment.

[0003] Currently, simulation testing of feeder automation strategies mainly employs pure digital real-time simulation or traditional centralized hardware-in-the-loop testing (HIL) schemes. While pure digital simulation is flexible and efficient, it typically simplifies or ignores the dynamic response characteristics and communication interaction processes of real physical terminals, making it difficult to reflect the timing and deviations of the distribution terminal hardware's actions under actual signal excitation. Centralized HIL testing, on the other hand, connects terminals to the same simulation platform, focusing on local closed-loop verification and lacking simulation of key influencing factors such as multi-party collaboration in the cloud-edge architecture, network transmission conditions, and dynamic changes in computing resources. When these conventional testing methods are directly used to verify cloud-edge collaborative feeder automation strategies, they often fail to reproduce the coordination errors and strategy execution defects caused by timing misalignments, interaction delays, and fluctuations in edge computing load between the cloud and the edge.

[0004] The main reason for these limitations is the current lack of a testing mechanism that can deeply integrate cloud-based digital master simulation with real power distribution terminal hardware deployed on multiple edge nodes in the loop. There is also a lack of effective means to dynamically perceive and collaboratively compensate for spatiotemporal inconsistencies between the cloud and edge during the simulation process, resource contention, and communication constraints. This results in significant differences between the testing environment and the actual cloud-edge collaborative operating environment, leading to insufficient confidence in the verification results and making it difficult to support the performance evaluation and optimization iteration of feeder automation strategies under complex cloud-edge operating conditions.

[0005] Therefore, how to provide a hybrid simulation testing method and platform that can reproduce the behavior of the entire feeder automation process under cloud-edge collaboration with high fidelity, and effectively cope with the impact of heterogeneous edge nodes and asynchronous communication and clock, thereby improving test fidelity and policy verification level, has become an urgent technical problem to be solved. Summary of the Invention

[0006] This application provides a cloud-edge collaborative feeder automation hybrid simulation test method and platform, which at least addresses the problems existing in the prior art.

[0007] A first aspect of this application provides a cloud-edge collaborative feeder automation hybrid simulation testing method, comprising the following steps: S1. Based on the automation strategy of the feeder under test and the topology of the distribution line, a cloud-edge hybrid simulation model is constructed. The hybrid simulation model includes a cloud-to-cloud feeder digital twin master simulation model and a hardware-in-the-loop simulation container for distribution terminals deployed on each edge node. S2. Analyze the control logic of the automation strategy of the feeder under test, divide the feeder topology into a cloud-based centralized decision-making simulation area and at least one edge-based local control simulation area, and generate a cloud-edge collaborative simulation task sequence. S3. For each edge local control simulation area, obtain the real-time computing resources and physical I / O interface status of the corresponding edge node, and dynamically determine the simulation step size, data interface mapping table and simulation clock reference of the edge node. S4. After triggering the hybrid simulation, according to the cloud-edge collaborative simulation task sequence, the simulation initialization parameters and the fault scenario timestamp sequence containing the fault occurrence time and fault type are sent to each edge node, and the global simulation clock is started. S5. Control each edge node to drive the power distribution terminal hardware to run in the hardware-in-the-loop simulation container according to the received simulation initialization parameters and fault scenario timestamp sequence, collect terminal protection action signals and electrical analog quantities in real time, and receive time-stamped data fed back by each edge node after adding edge timestamps. S6. Use the received time-stamped feedback data to perform spatiotemporal alignment with the simulation state of the main simulation model in the cloud at the corresponding time, identify data gaps and timing misalignments, use a compensation and correction algorithm based on reference vector interpolation to adjust the synchronization of the cloud-edge simulation state, and control the global simulation clock to advance according to the slowest edge node. S7. When the preset simulation end conditions are met, stop the hybrid simulation. Based on the full-process state variables recorded in the cloud and the sequence of terminal action events fed back by each edge node, evaluate the fault location accuracy, isolation coordination and power restoration time of the feeder automation strategy under cloud-edge collaboration, and generate a test report.

[0008] This application constructs a hybrid architecture of cloud-based digital twin master simulation and multi-edge hardware-in-the-loop simulation, and forms a collaborative task sequence based on policy logic partitioning, enabling the simulation to realistically reflect the interaction process between cloud-based decision-making and edge-based local control. Based on the real-time capabilities of edge nodes, the simulation step size is dynamically mapped, hardware response is triggered by fault timestamps, and reference vector interpolation compensation and synchronous advancement mechanisms are used to effectively and realistically reproduce the timing errors, resource competition and communication constraints in cloud-edge interaction, significantly improving test fidelity and the confidence of feeder automation strategy verification.

[0009] In this embodiment of the application, step S3, which dynamically determines the simulation step size, data interface mapping table, and simulation clock reference of the edge node, specifically includes: S31. Obtain the CPU utilization, memory availability, and I / O interrupt response latency of the edge node, and calculate the real-time simulation capability index of the node. S32. Based on the real-time simulation capability index and the electrical transient simulation accuracy required for the local control simulation area at the edge, set the minimum simulation step size and generate the electrical quantity-physical I / O channel mapping table of the power distribution terminal hardware terminals under the simulation step size. S33. Based on the cloud-based global simulation clock, synchronize the clock of the edge node and generate the local simulation propagation clock reference of the edge node on the global clock scale according to the set simulation step size.

[0010] This application utilizes the real-time computing resources and I / O status of edge nodes to quantitatively evaluate their simulation capabilities, and adaptively matches the simulation step size and interface mapping with the transient accuracy requirements of the region, avoiding resource waste or insufficient accuracy caused by fixed step size; at the same time, it generates local clock references for each edge node based on the cloud clock, ensuring the consistency and collaborativeness of the entire hybrid simulation clock system.

[0011] In this embodiment of the application, step S6 uses a compensation and correction algorithm based on reference vector interpolation to adjust the cloud edge simulation state synchronization, specifically including: S61. The node voltage amplitude and phase angle at the current moment of the cloud-based main simulation model are used as reference vectors and compared with the node electrical quantities with edge time scales fed back by the edge nodes at the same moment to form an observation vector, and the timing misalignment is detected. S62. If the timing misalignment exceeds the set threshold, construct the electrical quantity difference vector between adjacent step sizes of the edge nodes in the cloud, and perform linear interpolation compensation on the observation vector to generate the corrected cloud equivalent input. S63. Update the equivalent interface boundary conditions of the corresponding edge simulation region in the cloud master simulation model using the corrected equivalent input, drive the cloud simulation state correction, and send the corrected reference vector to the corresponding edge node for the edge node to use for the next long initial value correction.

[0012] This application establishes timing misalignment detection by comparing the reference vector and the observation vector, and uses the adjacent step size difference vector for linear interpolation compensation, which can effectively correct cloud edge state deviations caused by communication delays and clock asynchrony, and avoid simulation inaccuracies caused by the accumulation of deviations; at the same time, the corrected reference vector is fed back to the edge nodes for bidirectional correction, realizing closed-loop spatiotemporal synchronization and further improving the accuracy of collaborative simulation.

[0013] In this embodiment of the application, step S4, which involves sending a fault scene timestamp sequence containing the fault occurrence time and fault type to each edge node, includes: S41. Based on the fault test cases of the automated strategy of the feeder under test in the cloud, generate a sequence of fault events marked with a unified absolute timestamp for the entire system. Each fault event includes the fault type, fault location identifier and duration. S42. Based on the division of the cloud-edge collaborative simulation task sequence, the fault event sequence is split into local fault timestamp sub-sequences that only involve the corresponding edge local control simulation area. S43. Along with the simulation initialization parameters, the local fault timestamp subsequence is sent to the corresponding edge nodes, while the global fault event sequence containing all edge regions is retained in the cloud-based main simulation model for the main simulation to advance.

[0014] This application splits fault events with a unified timestamp across the entire system into local sub-sequences by region and distributes them accordingly. This allows each edge node to respond only to fault events related to its own region, effectively reducing unnecessary fault data parsing and computational load on the edge side and enhancing the edge's real-time simulation capabilities. Meanwhile, the cloud retains the global sequence, ensuring that the main simulation can fully reproduce the system-level fault evolution, achieving a balance between local refinement and global integrity.

[0015] In this embodiment of the application, step S5, which controls each edge node to drive the power distribution terminal hardware within the hardware-in-the-loop simulation container, collects terminal protection action signals and electrical analog quantities in real time, and receives time-stamped data fed back by each edge node after adding edge time stamps, specifically includes: S51. Create a real-time thread at the edge node that matches the simulation step size, and iteratively solve the transient equations of the power system in the hardware-in-the-loop simulation container under the drive of the local simulation advancement clock reference. S52. When the local simulation clock advances to the fault occurrence time in the received fault scenario timestamp sequence, the corresponding fault electrical analog quantity is injected into the power distribution terminal hardware through the I / O interface, and the terminal's input and output signals are collected. S53. The edge node combines the collected terminal action signal with the local simulation electrical quantity at the current time, and adds a double time stamp containing the local simulation clock time and the actual physical time to form time-stamped feedback data.

[0016] This application ensures that the electrical signals injected by the hardware in the loop strictly conform to the dynamic characteristics of the power system by iteratively solving the transient equations with real-time thread and step size matching; it accurately injects and collects terminal responses at the time of fault, and records the local simulation time and physical time with dual time scales, providing a time reference benchmark with controllable accuracy for subsequent spatiotemporal alignment in the cloud, which can accurately reflect the action latency of the real physical terminal under cloud-edge collaborative conditions.

[0017] In this embodiment of the application, step S6, which controls the global simulation clock to advance according to the slowest edge node, specifically includes: S64: Set a global simulation step counter in the cloud and collect the latest local simulation clock time from all edge nodes at each global step time; S65. Identify the minimum advance time among all local simulation clock times, adjust the global simulation clock to the minimum advance time, and determine whether there is a continuous delay in the lagging edge nodes. S66. If the delay of multiple consecutive steps of an edge node exceeds the preset tolerance threshold, the cloud-based main simulation model will perform model reduction processing on the simulation area corresponding to the edge node, convert the power distribution lines of the area into a lumped impedance model, and increase the calculation step size to accelerate its synchronization.

[0018] This application uses the slowest edge node as a reference to advance the global clock, which can prevent faster computing nodes from leading the hardware response and causing data mismatch, while avoiding global blocking caused by the continuous delay of individual nodes. By automatically identifying the delay and reducing the model order to increase the calculation step size, it can intelligently achieve a dynamic balance between fidelity and advancement efficiency, ensuring stable operation of hybrid simulation for a long time.

[0019] In this embodiment of the application, step S7, which evaluates the fault location accuracy, isolation coordination, and power restoration time of the feeder automation strategy under cloud-edge collaboration, specifically includes: S71. Extract the overcurrent and undervoltage signals of each switch before and after the fault moment from the cloud-based full-process state variables, and perform causal correlation with the terminal action event sequence fed back by each edge node to calculate the fault location accuracy. S72. Analyze the time difference between the protection actions and reclosing coordination of adjacent switches, combine the recorded cloud-edge interaction delay, evaluate the coordination of isolation actions, and distinguish the contribution ratio of cloud decision delay and edge local control delay. S73. Calculate the time from the occurrence of the fault to the restoration of power supply to the non-faulty section, and subtract the known inherent delay of I / O response and communication delay in the hybrid simulation to obtain the net power restoration time index.

[0020] This application performs causal correlation analysis based on the global state in the cloud and the actual action records of the edge terminal, which can accurately locate the performance of the strategy at each stage of fault handling; it separates the cloud-edge interaction latency from the action coordination and quantifies the latency contribution of the cloud and the edge respectively, providing a fine quantitative basis for optimizing the cloud-edge function division of the strategy; the net recovery time index eliminates the latency of the simulation platform itself and objectively reflects the power recovery efficiency of the strategy itself.

[0021] In this embodiment of the application, after generating the electrical quantity-physical I / O channel mapping table in step S32, the method further includes: The control edge node outputs standard test excitation to the power distribution terminal hardware input terminal and collects the hardware output response. By comparing the excitation and response, the mapping coefficient and nonlinear deviation between electrical quantities and physical I / O channels are automatically calibrated, and the calibration results are updated to the data interface mapping table.

[0022] This application automatically completes the mapping calibration through the actual hardware stimulus response before the simulation begins, which can eliminate individual differences and nonlinear errors caused by different terminal hardware, cables and interface circuits, ensure the accurate transmission between electrical quantities and physical signals in subsequent hybrid simulations, and improve the testing accuracy of the entire hardware in the loop.

[0023] This application embodiment also includes: S8. During the hybrid simulation operation, monitor the computing resource utilization and cloud-edge communication round-trip latency of each edge node in real time. S9. When the resource utilization rate of any edge node is lower than the first threshold or the communication latency exceeds the second threshold, the simulation tasks corresponding to the non-critical local control simulation area undertaken by the edge node are migrated to the main simulation model in the cloud, and the area is transformed into a simplified dynamic equivalent model. S10. Update the cloud-edge collaborative simulation task sequence and the corresponding edge node data interface mapping table, reissue the updated simulation initialization parameters to each edge node, and perform clock synchronization between the cloud and edge nodes to rebalance the load between the cloud and edge nodes before continuing the hybrid simulation.

[0024] This application dynamically senses changes in edge node resources and network degradation during simulation operation, promptly migrates non-critical simulation tasks to the cloud and simplifies them accordingly, and avoids overall simulation distortion or interruption due to limited local edge resources or communication deterioration. This enables the hybrid simulation system to have adaptive load rebalancing capabilities, significantly improving the robustness and success rate of long-term simulation tasks.

[0025] In this embodiment of the application, after updating the equivalent interface boundary conditions of the corresponding edge simulation region in the cloud-based main simulation model using the corrected equivalent input in step S63, the method further includes: S631. Collect the corrected equivalent input sequence generated in multiple consecutive synchronization cycles in the cloud, and construct the time series samples of the boundary state quantities of each edge simulation region. S632. Using a pre-trained time-series prediction model, predict the boundary state trend of the edge simulation region in the next synchronization cycle based on time-series samples, and form the predicted boundary conditions. S633. Before the next synchronization cycle arrives, the predicted boundary conditions are sent to the corresponding edge nodes in advance, so that the edge nodes can use them as the initial values ​​of the local boundary for transient solution when they have not received the next reference vector from the cloud, and complete the correction after receiving the formal reference vector.

[0026] This application introduces a boundary state prediction and pre-deployment mechanism based on historical correction sequences, enabling edge nodes to perform local pre-calculations in advance while waiting for cloud synchronization vectors. This significantly reduces simulation stagnation caused by communication waiting, effectively compresses the synchronization idle time of cloud-edge collaborative simulation, and improves the overall efficiency of hybrid simulation. At the same time, subsequent correction of the formal reference vector ensures that the simulation accuracy is not lost.

[0027] In this embodiment of the application, after injecting the corresponding fault electrical analog quantity into the power distribution terminal hardware through the I / O interface and starting to collect the terminal's input and output signals in step S52, the method further includes: S521. When the edge node detects that the terminal input / output signal is inconsistent with the expected protection action logic of the local simulation within the preset response time window, record the simulation state snapshot at the current moment. S522. Keep the local simulation clock from advancing. Based on the inconsistency type, iteratively correct the injected faulty electrical analog parameters within the preset adjustment range, and re-drive the power distribution terminal hardware until the terminal action signal is consistent with the expected logic or the maximum number of iterations is reached. S523. Update the fault electrical simulation quantity according to the correction parameters at the end of the iteration, continue to advance the local simulation clock, and feed back the iteration process and correction parameters along with the time stamp data to the cloud.

[0028] This application achieves real-time closed-loop consistency verification of fault injection and terminal actions by pausing the local clock and iteratively correcting fault quantities. It can quickly locate and correct inconsistencies in actions caused by simulation model deviations or interface parameter errors, improve the accuracy of hardware-in-the-loop simulation and the reliability of fault inversion, and limit the impact on the global simulation clock to a controllable delay range.

[0029] In this embodiment of the application, after generating the test report in step S7, the method further includes: S71. Extract fault test cases that fail to meet the preset collaborative performance indicators from the test report. Analyze the key control parameters that cause performance deviations for the fault location accuracy, isolation coordination, and power restoration time indicators in the extracted test cases. S72. Based on the control logic of the automation strategy of the feeder under test, an adjustment vector for key control parameters is generated, and the control parameters of the cloud-based centralized decision-making simulation area and the edge-based local control simulation area are modified according to the adjustment vector. S73. Update the cloud-edge collaborative simulation task sequence and fault scenario timestamp sequence according to the modified control parameters, re-trigger the hybrid simulation, and perform regression verification on the adjusted strategy until the collaborative performance indicators of all fault test cases meet the requirements.

[0030] This application establishes a closed-loop iterative process from testing and evaluation to automatic optimization of strategy parameters. It can automatically identify strategy defects based on hybrid simulation results and generate targeted parameter adjustment schemes, driving a new round of collaborative simulation verification. This significantly improves the optimization efficiency and performance compliance rate of feeder automation strategies under cloud-edge collaborative conditions, and realizes the integration of testing and optimization.

[0031] A second aspect of this application provides a cloud-edge collaborative automated hybrid simulation test platform for feeders, including a cloud server and multiple edge simulation nodes; The cloud server is equipped with a main simulation module and a collaborative management module, while the edge simulation nodes are equipped with a hardware-in-the-loop simulation module and an interface adaptation module. The main simulation module is used to build and run the cloud-based feeder digital twin main simulation model; The collaborative management module is used to parse the control logic partition of the automation strategy of the feeder under test, generate the cloud-edge collaborative simulation task sequence, dynamically determine the simulation step size, data interface mapping table and simulation clock reference of each edge simulation node, and execute the collaborative control and evaluation functions in the above method. The hardware-in-the-loop simulation module is used to run the hardware-in-the-loop simulation container of the power distribution terminal on the corresponding edge simulation node, drive the power distribution terminal hardware and acquire signals; The interface adapter module is used to realize physical I / O connection and signal mapping with the power distribution terminal hardware.

[0032] The platform provided in this application constructs a hybrid testing system that deeply integrates cloud-based digital twin main simulation and multi-edge power distribution terminal hardware in the loop through the collaboration between the main simulation module of the cloud server and the semi-physical simulation module of the edge simulation node. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating a cloud-edge collaborative hybrid simulation testing method for feeder automation, provided as an embodiment of this application. Detailed Implementation

[0034] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0035] In the field of distribution network feeder automation, to ensure rapid fault location, isolation, and power restoration, simulation testing and verification of feeder automation strategies are often required in engineering projects. Existing testing methods typically rely on two approaches: one is pure digital simulation, which involves building a distribution network model and protection control logic model entirely in a computer for offline simulation; the other is centralized hardware-in-the-loop simulation, which connects one or a small number of distribution terminals to a single simulation host through a dedicated interface to verify the actual response of the terminals. However, with the promotion of edge computing and cloud-edge collaborative architecture in distribution networks, feeder automation strategies are increasingly being deployed in a fragmented manner. The cloud undertakes global topology analysis, power flow calculation, and centralized recovery decision-making, while each edge node performs rapid protection, distributed islanding identification, and local control locally. This collaborative model of "centralized decision-making in the cloud and local control at the edge" means that the actual feeder automation behavior depends not only on the control logic of each node itself, but also is profoundly affected by factors such as heterogeneous computing resources of edge nodes, communication latency jitter, asynchronous cloud-edge clocks, and timing coupling caused by partitioning of control logic.

[0036] In practice, it has been found that pure digital simulation cannot incorporate the time response characteristics and physical input / output delays of real power distribution terminal hardware, nor can it accurately reflect the inconsistency in simulation step size caused by limited computing resources at edge nodes. While centralized hardware-in-the-loop solutions connect a small number of real terminals, their simulation core remains concentrated on a single host, lacking interactive simulation with the heterogeneous computing environment of multiple edge nodes, cloud-edge communication networks, and real physical constraints such as clock asynchrony. This makes it impossible to reproduce the coordination errors and protection strategy conflicts caused by asynchronous progress of edge nodes. This indicates that existing testing methods cannot realistically reflect the actual performance of feeder automation strategies after cloud-edge collaborative deployment in a simulation environment, thus reducing the confidence of test results in providing guidance for field operations, constituting a technical problem that urgently needs to be solved.

[0037] The root cause of this problem lies in the lack of a hybrid simulation architecture that deeply integrates cloud-based digital master simulation with hardware-in-the-loop simulation of power distribution terminals distributed across multiple edge nodes. Furthermore, it lacks a compensation and synchronization control mechanism for spatiotemporal inconsistencies during cloud-edge collaborative simulation within this architecture. Simply splicing cloud simulation and edge hardware-in-the-loop simulation together, with each edge node advancing its simulation based on its own computing power and independent clock, inevitably leads to differences in simulation progress between edge nodes and state misalignment between the edge and the cloud. Data fed back from the edge to the cloud is scattered and inconsistent in the temporal dimension, making it difficult to align with a unified global timing sequence for collaborative strategy evaluation. Therefore, achieving high-fidelity cloud-edge collaborative testing necessitates a testing method that can dynamically sense the real-time capabilities of edge nodes, coordinate simulation step size and clock reference, and perform online compensation and correction without interrupting the simulation when timing misalignments occur.

[0038] Based on the above understanding, this application proposes a cloud-edge collaborative hybrid simulation testing method for feeder automation. In the analysis, the cloud-edge collaborative simulation testing is abstracted into three key actions: First, the partitioning mapping of the strategy and the quantitative description of the edge node capabilities, i.e., dividing the feeder topology into regions according to the strategy under test, and acquiring the real-time computing resources and physical I / O status of edge nodes to dynamically determine the simulation step size and interface mapping of each edge region; second, constructing a cloud-edge temporal semantic consistency fault injection and data exchange mechanism, using fault scenario timestamp sequences and feedback data with edge timestamps to achieve cloud-edge event alignment; third, identifying and resolving the inherent spatiotemporal inconsistency problem in hybrid simulation, i.e., using a reference vector interpolation compensation correction algorithm to adjust the synchronization of cloud-edge simulation states, and advancing the global clock according to the slowest edge node to avoid causal disorder. This leads to a closed-loop testing process: "modeling and partitioning—capability-aware step-size mapping—fault timestamp collaborative injection—hardware-in-the-loop acquisition with dual time-stamped feedback—spatiotemporal alignment compensation and slowest node synchronization—collaborative index evaluation." This enables the simulation environment to realistically reproduce the timing discrepancies, resource contention, and communication constraints introduced by cloud-edge interaction, significantly improving the fidelity of feeder automation strategy verification and the guidance of test results for engineering deployment.

[0039] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1 The following is an explanation using specific examples.

[0040] Please refer to Figure 1 , Figure 1 The first aspect of this application provides a cloud-edge collaborative feeder automation hybrid simulation testing method, comprising the following steps: S1. Based on the automation strategy of the feeder under test and the topology of the distribution line, a cloud-edge hybrid simulation model is constructed. The hybrid simulation model includes a cloud-to-cloud feeder digital twin master simulation model and a hardware-in-the-loop simulation container for distribution terminals deployed on each edge node. S2. Analyze the control logic of the automation strategy of the feeder under test, divide the feeder topology into a cloud-based centralized decision-making simulation area and at least one edge-based local control simulation area, and generate a cloud-edge collaborative simulation task sequence. S3. For each edge local control simulation area, obtain the real-time computing resources and physical I / O interface status of the corresponding edge node, and dynamically determine the simulation step size, data interface mapping table and simulation clock reference of the edge node. S4. After triggering the hybrid simulation, according to the cloud-edge collaborative simulation task sequence, the simulation initialization parameters and the fault scenario timestamp sequence containing the fault occurrence time and fault type are sent to each edge node, and the global simulation clock is started. S5. Control each edge node to drive the power distribution terminal hardware to run in the hardware-in-the-loop simulation container according to the received simulation initialization parameters and fault scenario timestamp sequence, collect terminal protection action signals and electrical analog quantities in real time, and receive time-stamped data fed back by each edge node after adding edge timestamps. S6. Use the received time-stamped feedback data to perform spatiotemporal alignment with the simulation state of the main simulation model in the cloud at the corresponding time, identify data gaps and timing misalignments, use a compensation and correction algorithm based on reference vector interpolation to adjust the synchronization of the cloud-edge simulation state, and control the global simulation clock to advance according to the slowest edge node. S7. When the preset simulation end conditions are met, stop the hybrid simulation. Based on the full-process state variables recorded in the cloud and the sequence of terminal action events fed back by each edge node, evaluate the fault location accuracy, isolation coordination and power restoration time of the feeder automation strategy under cloud-edge collaboration, and generate a test report.

[0041] In this embodiment, in S1, the automation strategy of the feeder under test can be understood as the complete set of control logic for protection, isolation, and power restoration that needs to be verified when a fault occurs in the distribution network. This includes, for example, overcurrent protection parameter setting schemes, reclosing coordination timing, and power transfer path selection rules. The distribution line topology refers to the electrical connection diagram between switches, segments, branches, and distribution transformers in the feeder section, which can be represented by a node-branch model. The cloud-based feeder digital twin master simulation model is a digital model running on a cloud server, mirroring the dynamic behavior of the real feeder. It can perform real-time or ultra-real-time calculations based on electromagnetic transient or electromechanical transient equations, outputting the amplitude and phase angle of voltage and current at each node. The distribution terminal hardware-in-the-loop simulation container is a hardware and software operating environment deployed on edge nodes. It connects the distribution terminal entity to the simulation loop through physical input / output interfaces, enabling the terminal hardware to collect electrical quantities and issue action signals during the simulation process, just as it would in an actual distribution network. Edge nodes can be devices with computing and communication capabilities deployed on the side of substations, switching stations, or line towers, such as edge computing gateways or industrial control computers.

[0042] In S2, analyzing the control logic of the feeder automation strategy under test refers to decomposing the strategy code or configuration parameters from an overall functional perspective into decisions that fall within the entire system scope and require information from multiple terminals, and actions that can be executed based solely on the local electrical quantities and states of the node. The cloud-centralized decision-making simulation area refers to the parts of the feeder topology where the cloud performs protection coordination judgments, power transfer path searches, and recovery strategy decisions based on global information, such as the power transfer selection logic of the tie switch. The edge-local control simulation area refers to the parts of the feeder topology where a single edge node directly completes protection output, reclosing, and distributed islanding judgments based on information collected from local terminals, such as the instantaneous overcurrent protection and reclosing of the feeder output circuit breaker. The cloud-edge collaborative simulation task sequence is an arrangement table of the simulation calculations, hardware drivers, and data interaction steps required by the cloud and each edge node throughout the hybrid simulation process, organized chronologically or sequentially by event. It can include the executing entity, triggering conditions, and estimated duration of each step.

[0043] In S3, real-time computing resources refer to the computing power indicators such as CPU computing power, memory capacity, and hard disk or flash memory read / write bandwidth currently available to the edge node. Physical I / O interface status refers to the available number, signal type, and current connection integrity of analog and digital input / output channels connected between the edge node and the power distribution terminal hardware. The simulation step size is the time interval between the edge node solving the transient equations of a primary power system and updating the results to the output interface; it can be understood as the time span between two adjacent simulation state updates. The data interface mapping table is a form specifying which electrical node voltages and branch currents within the simulation container are converted into signals perceptible to the terminal hardware through which physical I / O channels, with what scaling factor and linear or nonlinear relationship. For example, mapping the simulated value of phase A voltage to analog output channel 1 according to the transformation ratio. The simulation clock reference is the origin and interval scale of the time coordinate for the edge node to advance the local simulation; it is aligned with the global simulation clock in the cloud to ensure time comparability of cloud-edge events.

[0044] In S4, the fault scenario timestamp sequence is a set of fault event records ordered by time. Each record clearly indicates the fault occurrence time, fault type, and fault location identifier. The fault occurrence time uses an absolute time value calculated from the global simulation clock startup time. Fault types can include single-phase grounding, two-phase short circuit, two-phase ground short circuit, and three-phase short circuit, etc. The fault location identifier indicates which branch, line segment, or node of the feeder topology the fault occurred near. For example, the fault scenario timestamp sequence can be formed into a time-event table, and the cloud and each edge node inject the corresponding fault electrical quantities at the corresponding time according to the table. Simulation initialization parameters include the initial power flow state of each area simulation model, the initial switching position, terminal setting parameters, and load power setting values, etc.

[0045] In S5, terminal protection action signals refer to the switching signals generated by the distribution terminal when it detects overcurrent, undervoltage, or frequency abnormalities, such as trip outputs, reclosing initiation, and lockout. Electrical analog quantities refer to the instantaneous values ​​or fundamental RMS values ​​of voltage and current waveforms acquired by the terminal through current transformers and analog-to-digital converters. Adding edge time stamps means that when edge nodes generate feedback data, they append their own local simulation clock time and the physical time read from the hardware real-time clock to each data set, thus forming a traceable time source marker. Time-stamped data is a data packet containing the above time stamps, terminal action signals, and electrical analog quantities, which can be transmitted from the edge node to the cloud via the communication network.

[0046] In S6, spatiotemporal alignment maps feedback data from different edge nodes to a unified timeline of the global simulation clock in the cloud, enabling a comparison between the simulation state at the same moment in the main simulation model in the cloud and the reported states of each edge node. Data gaps refer to the absence of feedback data from a particular edge node during certain time periods on the global timeline. Timing misalignment refers to a causal reversal or significant lag between the feedback data from an edge node at the same moment and the state in the cloud at the same moment. The reference vector interpolation compensation correction algorithm is a signal processing technique where the cloud uses its own calculated node voltage amplitude and phase angle as a reference to estimate the correction value by performing linear or nonlinear interpolation between adjacent step sizes on the lagging or missing edge feedback data. This correction value is then used to update the boundary conditions of the cloud-edge interface. Advancing according to the slowest edge node means that the global simulation clock does not advance according to the time of the fastest node, but instead waits and advances gradually according to the local time reached by the edge node with the slowest completion progress, preventing fast nodes from causally depending on future states.

[0047] In S7, the full-process state variables are continuous time sequences of node voltages, branch power, switch states, and terminal action flags stored in the cloud from the start to the end of the simulation. The terminal action event sequence consists of time-ordered records of protection startup, trip outputs, reclosing actions, and blocking events fed back from each edge node. Fault location accuracy measures the consistency between the fault section determined by the strategy and the actual set fault point; it can be evaluated by the proportion of correctly identified fault sections out of all tests. Isolation coordination refers to the reasonableness of the timing coordination between adjacent switch protection actions and reclosing, and whether phenomena such as cascading tripping, excessively large or small isolation ranges occur. Power restoration time is the time elapsed from the occurrence of a fault to the successful restoration of power to the non-faulty section via tie switch transfer or upstream reclosing.

[0048] In a real-world cloud-edge collaborative environment, the testing challenge of feeder automation strategies lies in the fact that the strategy itself is split into two parts: global decision-making in the cloud and local control at the edge. These two parts operate on physically separate and heterogeneous performance platforms. S1 first establishes a hybrid simulation model structure corresponding to the strategy deployment method, allowing both the cloud and edge to have simulation environments reflecting their own behaviors. S2, building on this, divides the strategy's control logic into regions according to the actual deployment boundaries and orchestrates task sequences. This ensures that the spatial allocation of simulation tasks is consistent with the logical allocation of the real cloud-edge architecture, avoiding misalignment between the simulation environment and the target deployment environment at the control boundaries. S3's dynamic capability awareness and step size mapping, building on S1 and S2, solves the problem of inconsistent simulation rhythms caused by performance differences among edge nodes. If all edge nodes are forced to run at the same time step, nodes with weaker computing power may fail to complete calculations in time, leading to simulation distortion or timeouts. By acquiring real-time computing resources and I / O status and setting their respective simulation step size and clock reference accordingly, different edge nodes can perform asynchronous calculations according to their capabilities while maintaining correspondence with global time through the clock reference. This establishes an adaptable basic framework for time coordination across the entire system before simulation begins. S4 issues a fault scenario timestamp sequence and starts the global clock, providing a unified event-driven source for the entire hybrid simulation after the aforementioned partitioning and step size settings are completed. The core of cloud-edge collaborative testing is to examine whether the interaction and coordination between cloud decision-making and edge local control are correct after a fault occurs. If a unified timestamp sequence is not used and each node generates faults independently, the temporal relationship of fault events between different nodes may deviate, causing the action timing coordination analysis to lose a unified time reference and failing to truly reflect the coordinated performance of the strategy. Therefore, S4 establishes a consistent event timeline between the cloud and the edge based on S2 and S3. S5 drives the hardware to run in the loop and feeds back time-stamped data with edge timestamps. Under the unified event triggering of S4, it truly utilizes the power distribution terminal hardware to respond to the fault in the simulation loop, while recording the timestamp of the actual response at the edge. If only the terminal action signals are collected without timestamps, the cloud cannot determine the exact position of these actions on the global timeline after receiving them, and subsequent timing coordination analysis is impossible. Edge time stamps encompass both local simulation times and physical times, mapping simulation logic time to actual hardware response time and providing necessary time markers for S6 spatiotemporal alignment. S6's spatiotemporal alignment and compensation synchronization unifies the scattered feedback data from S5 with timing differences onto the global timeline and interpolates and compensates for delayed or missing data. This allows the cloud-based master simulation model to synthesize approximately consistent time-section states and edge feedback, driving the global clock to advance based on the slowest node.This step connects the time-consistency processing between the distributed edge feedback in S5 and the overall evaluation in S7. This ensures that although each edge node operates asynchronously, the entire system maintains causal order, avoiding logically inexplicable causal contradictions caused by individual nodes prematurely generating advanced states. Upon completion of the simulation, S7 utilizes the time-aligned state variables and terminal action event sequences processed in S6 to evaluate the strategy's performance from three dimensions: fault location, isolation coordination, and power restoration time. Because S1 through S6 have placed the tested strategy in a dynamic environment close to real cloud-edge physical constraints, the evaluation metrics obtained in S7 reflect the actual collaborative effect of the strategy under the combined influence of computational heterogeneity, communication latency jitter, and asynchronous cloud-edge clocks, rather than just its performance in an ideal, purely logical environment. This makes the test report valuable for engineering deployment.

[0049] In some embodiments of this application, step S3, which dynamically determines the simulation step size, data interface mapping table, and simulation clock reference of the edge node, specifically includes: S31. Obtain the CPU utilization, memory availability, and I / O interrupt response latency of the edge node, and calculate the real-time simulation capability index of the node. S32. Based on the real-time simulation capability index and the electrical transient simulation accuracy required for the local control simulation area at the edge, set the minimum simulation step size and generate the electrical quantity-physical I / O channel mapping table of the power distribution terminal hardware terminals under the simulation step size. S33. Based on the cloud-based global simulation clock, synchronize the clock of the edge node and generate the local simulation propagation clock reference of the edge node on the global clock scale according to the set simulation step size.

[0050] In this embodiment of the application, steps S31 to S33 are further limitations on dynamically determining the simulation step size, data interface mapping table, and simulation clock reference.

[0051] In S31, CPU utilization refers to the proportion of clock cycles in which the edge node's central processing unit is in a non-idle state within a statistical period, reflecting the current computational load of the processor. Memory margin is the unused portion of the edge node's physical memory, affecting the available space for matrix operations and state variable storage in transient simulations. I / O interrupt response latency is the time interval between a change in external hardware signals and the start of execution of the edge node's operating system interrupt service routine, typically measured in microseconds or milliseconds. The real-time simulation capability index is a one-dimensional comprehensive evaluation value obtained by normalizing and weighting the above three indicators. It is used to quantitatively measure the potential of the edge node to support real-time simulation solutions in its current state. For example, weighting factors can be set so that the closer the CPU utilization is to full load, the smaller the memory margin, and the slower the I / O interrupt response, the lower the index value, indicating a weaker real-time simulation support capability of the node. Let the real-time simulation capability index be C (CapabilityIndex), with a value range of [0, 1]. The closer the value is to 1, the stronger the simulation capability; the closer the value is to 0, the weaker the simulation capability. It can be calculated using the following weighted product (or weighted sum) form: C = w_CPU C_CPU + w_Mem C_Mem + w_IO C_IO Here, w_CPU, w_Mem, and w_IO are the weighting coefficients for CPU utilization, memory availability, and I / O interrupt response, respectively, satisfying w_CPU + w_Mem + w_IO = 1. In a specific embodiment, w_CPU can be set to 0.5, w_Mem to 0.3, and w_IO to 0.2 to highlight the dominant role of CPU computing power in real-time simulation.

[0052] In S32, the electrical transient simulation accuracy refers to the maximum allowable deviation range between the result and the theoretical true value or reference waveform when performing numerical integration calculations on the high-frequency transient processes of voltage and current after a fault. It is typically characterized by the percentage of amplitude error and the angle of phase error. The minimum simulation step size is the maximum time step value that the numerical integration algorithm can use while meeting this accuracy requirement. A smaller step size results in a more refined reproduction of the transient waveform, but also requires a higher computational speed. For example, if an edge local control simulation area only needs to capture the effective value changes of the fundamental frequency component, the minimum simulation step size can be set to the tens to hundreds of microseconds level; if there are power electronic devices in the area that need to characterize high-frequency switching ripple, the minimum simulation step size may need to be in the sub-microsecond range. The electrical quantity-physical I / O channel mapping table of the distribution terminal hardware terminals defines which electrical node voltage or which branch current in the simulation model is output to the distribution terminal hardware through which analog output channel of the edge node, with what scaling factor and transformation relationship, and also defines which digital I / O channels the terminal input and output signals are fed back into the simulation container.

[0053] In S33, the cloud-based global simulation clock serves as the unified time coordinate system upon which the main cloud simulation model progresses, and it is also the reference for comparing all events in time within the hybrid simulation. Clock synchronization refers to aligning the local clocks of edge nodes with the cloud-based global simulation clock using network clock synchronization protocols such as NTP, Network Time Protocol, or PTP and Precision Time Protocol, eliminating or compensating for initial time discrepancies between the two. Generating a local simulation advancement clock reference on the global clock scale can be understood as generating a series of discrete advancement time points at the edge nodes, with the same starting time and time unit as the global clock, and using the local simulation step size as the interval. Each time an edge node reaches an advancement time point, it performs a transient calculation and I / O data exchange. In this way, even if the step sizes of each edge node are different, their advancement time points still have a transparent correspondence with the global time scale.

[0054] Overall, these three sub-steps, through a progressive arrangement from capability awareness, step size and mapping determination to clock synchronization, enable edge node simulation parameters to adaptively match their own real-time state and regional simulation requirements, and establish a timing benchmark comparable to the cloud. S31 first obtains the edge node's CPU utilization, memory availability, and I / O interrupt response latency, then calculates the real-time simulation capability index. This step transforms the dynamic load of the edge node into comparable and usable quantitative information, providing a basis for subsequent differentiated simulation parameter decisions and avoiding overload or wasted computing power that might result from applying the same synchronization size to all edge nodes. S32 then uses this index in conjunction with the electrical transient simulation accuracy requirements to set the minimum simulation step size, and generates an electrical quantity-physical I / O channel mapping table accordingly. This ensures that both the simulation step size and interface mapping dimensions are aligned with the actual capabilities of the node and the physical characteristics of the measured area, allowing the simulation container to solve at an appropriate rate while ensuring correct matching between electrical quantities and terminal hardware interfaces at the corresponding rate. S33 further synchronizes the edge nodes' clocks with the global simulation clock in the cloud as a reference, and divides the local advancement time points based on the previously determined simulation step size. This allows edge nodes with different advancement time sizes to be mapped onto the same global time axis. Subsequently, whether the cloud performs spatiotemporal alignment or slowest node advancement control, it can directly use these clock references for unified time position conversion. In this way, S31 to S33 form a layered progressive architecture of capability awareness—parameter determination—clock connection, which collaboratively enables each edge node to advance the simulation with appropriate accuracy without exceeding its own computing capabilities and to maintain synchronization with the global time, thereby supporting the timing consistency of the entire hybrid simulation and the accuracy of strategy evaluation.

[0055] In some embodiments of this application, step S6 employs a compensation and correction algorithm based on reference vector interpolation to adjust the cloud-edge simulation state synchronization, specifically including: S61. The node voltage amplitude and phase angle at the current moment of the cloud-based main simulation model are used as reference vectors and compared with the node electrical quantities with edge time scales fed back by the edge nodes at the same moment to form an observation vector, and the timing misalignment is detected. S62. If the timing misalignment exceeds the set threshold, construct the electrical quantity difference vector between adjacent step sizes of the edge nodes in the cloud, and perform linear interpolation compensation on the observation vector to generate the corrected cloud equivalent input. S63. Update the equivalent interface boundary conditions of the corresponding edge simulation region in the cloud master simulation model using the corrected equivalent input, drive the cloud simulation state correction, and send the corrected reference vector to the corresponding edge node for the edge node to use for the next long initial value correction.

[0056] In this embodiment, steps S61 to S63 are a further refinement of adjusting the cloud edge simulation state synchronization using a compensation and correction algorithm based on reference vector interpolation.

[0057] In S61, the reference vector refers to the vector composed of the voltage magnitude and phase angle of each relevant node calculated by the cloud-based main simulation model at the current global simulation clock time. It can be regarded as the state reference that the hybrid simulation should present under ideal timing. The observation vector refers to the vector constructed based on the electrical quantities of nodes with edge time scales fed back by edge nodes at the same time. Because edge nodes have computational delays, communication delays, and step size differences, the observation vector may correspond to an actual state that is not completely consistent with the current reference vector in the cloud in terms of time. The timing misalignment is a time deviation measure obtained by comparing the correspondence and numerical differences between the reference vector and the observation vector on the time axis. It can be characterized by the time offset value calculated by the angular difference in phase between the two or the slope of the waveform change. For example, the angle between the reference vector and the observation vector on the complex plane can be calculated, and then the angle difference can be converted into a time offset at the microsecond or millisecond level in combination with the current frequency.

[0058] In S62, a timing misalignment exceeding a set threshold indicates that the time deviation of the cloud-edge simulation state has reached a level that may affect the coordination and judgment of subsequent feeder automation strategies. The set threshold can be determined based on the tolerance range of the measured strategy for time errors; for example, it can be set as a fraction of the simulation step size or a fraction of the protection coordination time level. The electrical quantity difference vector is a vector composed of the electrical quantity changes calculated between two adjacent simulation steps at the edge node. It reflects the state change trend of the edge simulation region between adjacent steps. Linear interpolation compensation uses this difference vector to perform a linear extrapolation or internal extrapolation of the observed vector according to its distance from the target time point, thereby estimating the state value that the edge node should have been in without delay. The corrected cloud-edge equivalent input is the result of linear interpolation compensation.

[0059] In S63, equivalent interface boundary conditions are equivalent boundary parameters (voltage, current, or impedance) applied to the interface location representing a certain edge simulation region in the cloud-based main simulation model. These parameters replace the influence of the edge region on the global cloud model with equivalent values. Driving cloud simulation state correction refers to updating these boundary conditions with corrected equivalent inputs and then recalculating the global model in the cloud in one or more steps, transitioning the cloud state from a state inconsistent with the delay feedback to a new state consistent with the compensated state. The corrected reference vector is the corresponding node voltage magnitude and phase angle set obtained after cloud state correction. After being sent to the corresponding edge node, the edge node can use this reference vector to replace or correct its local initial state value at the start of its next simulation step, ensuring that subsequent local calculations start from a state aligned with the cloud state.

[0060] Overall, S61 to S63 form a closed loop of detection, compensation, updating, and feedback. S61 monitors the time deviation between the reference vector and the observed vector in real time, transforming the time synchronization problem between the cloud and the edge into a quantifiable time misalignment, providing a clear trigger condition for whether to initiate correction. When the misalignment exceeds the tolerable limit, S62 does not simply discard the observed vector or directly use its value. Instead, it constructs an electrical quantity difference vector between adjacent step sizes of the edge nodes and performs interpolation to pull the observed vector back to its timescale position near the reference vector. This provides real-time compensation for data lag caused by communication and computation delays in the digital domain, generating a corrected equivalent input to the cloud and avoiding disruption of the cloud state due to outdated feedback data. S63 converts the corrected equivalent input into cloud-edge interface boundary conditions and updates the cloud master simulation model, aligning the cloud's own state. The corrected reference vector is then sent back to the edge nodes as the initial anchor point for the next long-term calculation at the edge. In this way, the starting point of the next round of local solution at the edge node is proactively corrected to the state synchronized with the cloud, which helps to suppress the accumulation and amplification of timing misalignment in continuous simulation steps. These three sub-steps are interconnected, so that cloud-edge collaborative simulation does not need to stop and wait for a reset when timing asynchrony occurs, but instead continuously maintains the comparability and consistency of the entire system state through online compensation. This provides a state record that is closer to the real physical timing behavior of the cloud and edge when evaluating the collaborative indicators of subsequent feeder automation strategies.

[0061] In some embodiments of this application, step S4 involves sending a fault scene timestamp sequence containing the fault occurrence time and fault type to each edge node, including: S41. Based on the fault test cases of the automated strategy of the feeder under test in the cloud, generate a sequence of fault events marked with a unified absolute timestamp for the entire system. Each fault event includes the fault type, fault location identifier and duration. S42. Based on the division of the cloud-edge collaborative simulation task sequence, the fault event sequence is split into local fault timestamp sub-sequences that only involve the corresponding edge local control simulation area. S43. Along with the simulation initialization parameters, the local fault timestamp subsequence is sent to the corresponding edge nodes, while the global fault event sequence containing all edge regions is retained in the cloud-based main simulation model for the main simulation to advance.

[0062] In this embodiment of the application, steps S41 to S43 are a further refinement of sending a fault scene timestamp sequence containing the fault occurrence time and fault type to each edge node.

[0063] In S41, fault test cases are a set of pre-designed fault scenario descriptions to verify feeder automation strategies. These cases may include elements such as fault location, fault type, fault occurrence time, transition resistance magnitude, and fault duration. A system-wide unified absolute timestamp refers to a unique and consistent time stamp across the entire cloud-edge hybrid simulation system. It is calculated from the global simulation clock startup time and is unambiguous due to differences in physical clocks at the edge or cloud. A fault event sequence is a set of fault events arranged chronologically; each fault event can be viewed as a record with an occurrence time and attribute label. The fault location identifier indicates which branch, node, or segment of the feeder topology the fault occurred on, and can be in the form of node number, segment number, or latitude and longitude coordinates. The duration refers to the time elapsed from the fault's occurrence to its clearing, and can be used to simulate transient or permanent faults.

[0064] In S42, splitting refers to dividing the global fault event sequence generated in the cloud into multiple subsets based on the geographical or electrical coverage of each edge's local control simulation area, as determined in the cloud-edge collaborative simulation task sequence. Each subset retains only fault events related to the area managed by the corresponding edge node. The local fault timestamp subsequence can be understood as a sub-table containing only fault events that need to be responded to in the current edge area, sorted by timestamp. For example, if an edge area is only responsible for line protection between two switches in the middle of a feeder, then only events where the fault location falls within this line segment will be included in the local subsequence of that edge node, while fault events from other areas will be filtered out.

[0065] In S43, sending the simulation initialization parameters along with the local fault timestamp subsequence refers to packaging and sending it to the edge nodes along with other configuration parameters required for simulation startup, such as initial power flow and switch states. Retaining the global fault event sequence encompassing all edge regions means that no fault information is lost on the cloud side. The cloud still stores and drives its full-system main simulation based on the complete event sequence. Even if some fault events have been distributed to specific edge nodes, the cloud still needs these events to calculate the global power flow distribution and cross-regional decision coordination. Main simulation progression refers to the process of solving the cloud-based main simulation model along the time axis.

[0066] Overall, S41 to S43, through the coherent actions of generation, splitting, and distribution, ensures that edge nodes receive fault stimuli precisely corresponding to their control responsibilities, while avoiding the processing burden caused by receiving irrelevant fault information. Simultaneously, the cloud retains a global view to maintain consistency across the entire system simulation. S41 first generates a sequence of fault events with a unified absolute timestamp in the cloud based on fault test cases. This step transforms previously scattered fault assumptions into standard events that can be located on a global timeline, providing a unified time reference for cloud-edge collaborative event injection. Faults of different locations and types are defined within the same time coordinate system, and subsequent simulations, whether in the cloud or at the edge, operate based on the same time coordinate system. S42 then splits the faults according to the responsibility scope of each edge region, ensuring that each edge node only receives the local fault subsequence it needs to respond to. This on-demand distribution reduces the storage and processing overhead of irrelevant fault events for edge nodes and avoids logical ambiguity caused by receiving fault information beyond their decision-making scope. During the initialization phase, S43 distributes the split sub-sequences, while the cloud retains the complete event sequence. This allows the cloud to continuously calculate the state transitions of the entire feeder system based on global information, providing complete global event records to support subsequent evaluation of the timing of actions at each node. These three collaborative steps achieve the orderly generation, accurate distribution, and global supervision of fault events in the cloud-edge collaborative simulation environment. This lays the architectural foundation for consistent and well-distributed event data, enabling the local driving of real terminal responses at edge nodes and the unified evaluation of collaborative metrics in the cloud.

[0067] In some embodiments of this application, step S5 involves controlling each edge node to drive the power distribution terminal hardware within the hardware-in-the-loop simulation container, acquiring terminal protection action signals and electrical analog quantities in real time, and receiving time-stamped data fed back by each edge node after adding edge time stamps. Specifically, this includes: S51. Create a real-time thread at the edge node that matches the simulation step size, and iteratively solve the transient equations of the power system in the hardware-in-the-loop simulation container under the drive of the local simulation advancement clock reference. S52. When the local simulation clock advances to the fault occurrence time in the received fault scenario timestamp sequence, the corresponding fault electrical analog quantity is injected into the power distribution terminal hardware through the I / O interface, and the terminal's input and output signals are collected. S53. The edge node combines the collected terminal action signal with the local simulation electrical quantity at the current time, and adds a double time stamp containing the local simulation clock time and the actual physical time to form time-stamped feedback data.

[0068] In this embodiment, steps S51 to S53 are a further refinement of the edge node driving the power distribution terminal hardware to operate, collect signals and feed back time-stamped data within the hardware-in-the-loop simulation container.

[0069] In S51, real-time threads are task execution units with deterministic execution cycles, scheduled by the edge node operating system kernel. Their activation interval is strictly matched to the simulation step size. Scheduling jitter can be reduced by leveraging the operating system's real-time preemptive scheduling strategy or binding threads to dedicated processor cores. Driven by the local simulation propagation clock reference, this means that at each step of the local propagation time point sequence generated in S3 and aligned with the cloud global clock, a real-time thread execution iteration is triggered. Iteratively solving the power system transient equations means that each execution of the real-time thread performs a one-step numerical integration on the differential equations of the distribution network component models (such as lines, transformers, and loads) built within the simulation container, calculating a new state of voltage and branch current for each node. For example, the power system transient equations can be modeled using electromagnetic transient procedures or state-space node analysis methods, and the integration method can be the implicit trapezoidal rule or the backward Euler method.

[0070] In S52, fault electrical analog quantities are calculated based on the fault type and location, and should be injected into the hardware input channel of the distribution terminal to simulate the voltage and current waveform signals when a real fault occurs. For example, during a single-phase ground fault, the analog signal waveforms of the corresponding phase voltage decrease and the zero-sequence voltage increase are shown. Input signals refer to binary status quantities such as external switch position status, pressure plate activation / deactivation signals, and interlocking signals detected by the distribution terminal through the digital input circuit. Output signals refer to binary control quantities such as trip commands, closing commands, and alarm signals issued by the distribution terminal through the digital output circuit. I / O interfaces can include analog input / output channels such as voltage-type or current-type analog interfaces, digital input / output channels, and communication interfaces such as virtual I / O paths based on serial or Ethernet.

[0071] In S53, the combination of terminal action signals and local simulation electrical quantities refers to organizing the input and output switching signals acquired at the same time with the voltage and current analog quantities solved by the simulation container into a structured data record, so that the cloud can understand the electrical background in which the terminal action occurs. The local simulation clock time is the global time value corresponding to the current simulation step size recorded by the edge node, which adopts the local clock reference established on the global clock scale in S3. The actual physical time is the wall time read from the hardware real-time clock of the edge node itself, marking the physical time point when the data was acquired or time-stamped. The dual time stamp binds the simulation logic time and physical response time in the same data record, enabling the cloud to determine the chronological relationship of events in the simulation scenario and also distinguish the magnitude of physical delay.

[0072] Overall, S51 to S53, following the role of edge nodes in hybrid simulation, sequentially address three key issues: real-time calculation driving, fault synchronization injection and signal acquisition, and feedback data time stamping. S51 establishes a real-time thread matching the simulation step size, driving the edge node's simulation calculation process with a hardware timer, thus achieving predictable periodic execution behavior. The transient state of each calculation step provides a real-time updated electrical reference for subsequent signal injection and acquisition. S52 inserts a fault event-driven mechanism into this real-time calculation process. When the local simulation clock reaches a preset fault time, the corresponding fault electrical quantity is output to the actual power distribution terminal hardware input via the I / O channel, simultaneously capturing the terminal's input and output responses. This design makes the actual terminal operate as if triggered by fault electrical quantities in the actual power distribution network, establishing an excitation closed loop from simulation to hardware. S53 then combines and packages the terminal's actual protection action signals, real-time acquired electrical analog quantities, and dual-time-stamped information, ensuring that the data fed back to the cloud not only reports what actions the terminal performed but also marks the position of the action in the simulation logic time and the physical wall time. This provides traceable temporal dimension information for subsequent spatiotemporal alignment and global clock advancement in S6, enabling the cloud to identify the degree of temporal deviation between feedback data from different edge nodes and between feedback data and cloud status, and providing input for compensation and correction operations. These three progressive steps integrate the discrete edge simulation and physical terminal interaction process into an ordered, measurable time-series feedback mechanism, thereby supporting the real-time closed loop of the entire cloud-edge collaborative testing system.

[0073] In some embodiments of this application, step S6, which controls the global simulation clock to advance according to the slowest edge node, specifically includes: S64: Set a global simulation step counter in the cloud and collect the latest local simulation clock time from all edge nodes at each global step time; S65. Identify the minimum advance time among all local simulation clock times, adjust the global simulation clock to the minimum advance time, and determine whether there is a continuous delay in the lagging edge nodes. S66. If the delay of multiple consecutive steps of an edge node exceeds the preset tolerance threshold, the cloud-based main simulation model will perform model reduction processing on the simulation area corresponding to the edge node, convert the power distribution lines of the area into a lumped impedance model, and increase the calculation step size to accelerate its synchronization.

[0074] In this embodiment, steps S64 to S66 are specific limitations on controlling the global simulation clock to advance according to the slowest edge node.

[0075] In S64, the global simulation step counter is an incrementing count variable or time stamp variable maintained in the cloud. It increments by one step after each synchronization step of the entire system state is completed, tracking which synchronization cycle the hybrid simulation has reached. Each global step time refers to the time node where the synchronization cycle indicated by the global step counter ends and the next synchronization cycle begins. At this node, the cloud pauses its main simulation progression and enters the collection and judgment process. The latest local simulation clock time is the local simulation time value currently reached by each edge node, and this value increments forward as each edge node completes the calculation of a local step.

[0076] In S65, the minimum advance time is the minimum of the latest local simulation clock times reported by all edge nodes. It can be understood as the current time position of the slowest edge node. Adjusting the global simulation clock to this minimum advance time means that the cloud abandons its potentially advanced computation progress, regressing or waiting to reach this minimum value, using the time of the slowest node as the unified scale for the entire hybrid simulation system to move forward. Continuous latency refers to a situation where, within multiple adjacent global step cycles, the latest local simulation clock time reported by a certain edge node consistently lags behind other nodes; that is, this node repeatedly contributes to the minimum advance time, and the degree of lag is continuous.

[0077] In S66, the preset tolerance threshold is a pre-defined threshold for the number of consecutive delays or cumulative time deviations, used to define what delay states are acceptable temporary fluctuations and what delay states require intervention. Model reduction refers to simplifying a detailed and complex distribution line model, such as a distributed parameter model or multiple Π-type cascaded circuits, into an equivalent form with fewer parameters and lower computational complexity. The lumped impedance model uses lumped resistors, inductors, and capacitors to equivalently replace the distributed parameter characteristics of a section of the line, and its mathematical description is relatively simple. Increasing the computation step size means that after model reduction, due to changes in the model's time constant and dynamic characteristics, a larger integration step size can be used to solve the problem while maintaining numerical stability and the required accuracy, thereby reducing the computational cost per unit simulation time.

[0078] Overall, S64 to S66, through a progressive logic of collection, judgment, and intervention, evolves the control method of the global simulation clock from passively waiting for the slowest node to actively identifying and mitigating persistent delays. S64 collects the latest local clock progress of all edge nodes at each global step, aggregating the asynchronous simulation information originally scattered across nodes to the cloud, providing complete comparative data for system-wide progress synchronization decisions. S65 identifies the minimum value among these progress values ​​as the basis for global clock adjustment, establishing the progress of the slowest node as the time frontier that the entire system can reach together, avoiding causal contradictions caused by excessively rapid local advancements. Simultaneously, a continuous delay judgment mechanism distinguishes between occasional lags and persistent performance deficiencies, providing a basis for subsequent intervention decisions. When S66 identifies an edge node that is persistently lagging and exceeds the tolerance threshold, it does not simply discard the node or allow the overall simulation to be dragged down. Instead, it actively reduces the computational complexity of that region by reducing the model order and increasing the step size, enabling the node to accelerate its solution pace even with limited resources, gradually catching up with the global progress. These three sub-steps work together to maintain the causal consistency of the entire hybrid simulation system in the time dimension, and provide an adaptive recovery path for edge nodes whose performance degrades due to fluctuations in computing resources. This enables the entire cloud-edge collaborative simulation to maintain stable progress over a longer time scale and reduces the risk of global stagnation caused by short-term overload of individual nodes.

[0079] In some embodiments of this application, step S7, which evaluates the fault location accuracy, isolation coordination, and power restoration time of the feeder automation strategy under cloud-edge collaboration conditions, specifically includes: S71. Extract the overcurrent and undervoltage signals of each switch before and after the fault moment from the cloud-based full-process state variables, and perform causal correlation with the terminal action event sequence fed back by each edge node to calculate the fault location accuracy. S72. Analyze the time difference between the protection actions and reclosing coordination of adjacent switches, combine the recorded cloud-edge interaction delay, evaluate the coordination of isolation actions, and distinguish the contribution ratio of cloud decision delay and edge local control delay. S73. Calculate the time from the occurrence of the fault to the restoration of power supply to the non-faulty section, and subtract the known inherent delay of I / O response and communication delay in the hybrid simulation to obtain the net power restoration time index.

[0080] In this embodiment of the application, steps S71 to S73 are specific limitations on evaluating the fault location accuracy, isolation coordination, and power restoration time of the feeder automation strategy under cloud-edge collaboration.

[0081] In S71, an overcurrent signal refers to a situation where, during a short-circuit fault in the distribution network, the current amplitude flowing through the switch significantly exceeds the normal operating current limit within a short period. This signal can be collected and identified by the distribution terminal using a current transformer. An undervoltage signal refers to a situation where the voltage of a relevant node drops below a certain percentage of the rated voltage after a fault occurs. The terminal can use this signal to help determine the direction and segment of the fault. Causal correlation involves matching the electrical quantity exceeding limits recorded in the cloud-based state variables as the cause with the protection activation and tripping actions in the edge node feedback terminal action event sequence as the result. This determines whether each terminal responded in accordance with the preset logic of the strategy under the correct electrical disturbance excitation during a fault condition. Fault location accuracy can be understood as the proportion of fault segments identified by the strategy that match the actual set fault location in all set fault test cases. For example, it can be calculated by dividing the number of correctly identified fault segments by the total number of fault segments.

[0082] In S72, the time difference between adjacent switch protection actions and reclosing coordination refers to the interval between the tripping or reclosing times of the protection of two upstream and downstream switches under the same fault condition. This interval reflects whether the selectivity and tiered coordination of the upper and lower level protections are appropriate. Cloud-edge interaction delay is the one-way or round-trip communication time between the edge node sending time-stamped data to the cloud and the cloud issuing reference vectors or decision commands to the edge node. This can be recorded by the collaborative management module during hybrid simulation. The coordination evaluation of isolation actions focuses on whether the switches on both sides of the faulty section trip in a timely manner, whether there is any cascading tripping, and whether non-faulty sections are unnecessarily disconnected. Cloud decision delay refers to the logical processing time from receiving fault information uploaded from the edge to generating a global recovery or coordination decision in the cloud. Edge local control delay refers to the processing time from the edge node detecting an electrical quantity exceeding the limit to the local terminal generating a protection output action. The contribution ratio is a quantitative description of the proportion of these two delay times in the total coordination time difference.

[0083] In S73, non-faulty sections are the power supply areas on the feeder topology other than the section where the preset fault point is located. These areas should not lose power supply or should be restored via tie switches after a short power outage. The duration of power restoration is from the time of the fault occurrence on the global time axis until the voltage of the non-faulty section returns to normal levels and stabilizes. I / O response inherent latency is a fixed or calibrable lag caused by the physical characteristics of the power distribution terminal hardware in the hardware-in-the-loop simulation container, such as the analog signal acquisition and conversion latency and the action time of digital output relays. Communication latency is the network transmission time for time-stamped data from the edge node to the cloud. The net power restoration time index is an index value that more realistically reflects the recovery efficiency of the strategy itself, obtained by subtracting these known inherent latency and communication latency from the aforementioned duration.

[0084] Overall, S71 to S73 progressively evaluate the performance of feeder automation strategies under cloud-edge collaboration from three dimensions: the correctness of fault identification, the timing of multi-switch actions, and the timeliness of power restoration. S71 first establishes a causal relationship between the electrical state recorded in the cloud and the actual responses of each edge terminal, confirming from a global perspective whether each terminal was correctly triggered at the correct time under each fault. This lays the foundation for subsequent coordination quality and timeliness analysis at the event response level. Building on the causal relationship confirmed in S71, S72 further extracts the time difference between adjacent switch actions and introduces the decomposition of cloud-edge interaction delay. It decomposes the coordination quality into two traceable parts: cloud decision-making time and edge local control time. This decomposition allows the evaluation to not only indicate whether the coordination is harmonious but also distinguish whether the delay originates from cloud-edge communication and cloud computing or from the slow response of the edge terminal itself. S73 integrates the results of fault identification and isolation actions, statistically analyzing power restoration time from the perspective of power outage duration perceived by end users. It also separates inherent I / O and communication latency in the simulation system during the statistical analysis, yielding a net recovery time that more accurately reflects the merits of the policy algorithm itself. These three sub-steps work together to form a comprehensive evaluation system, from event localization to detailed analysis and final timeliness improvement. This ensures that the test report not only quantifies the accuracy of the policy but also provides diagnostically valuable detailed indicators for improving cloud-edge task partitioning, parameter tuning, and communication resource allocation.

[0085] In some embodiments of this application, after generating the electrical quantity-physical I / O channel mapping table in step S32, the method further includes: The control edge node outputs standard test excitation to the power distribution terminal hardware input terminal and collects the hardware output response. By comparing the excitation and response, the mapping coefficient and nonlinear deviation between electrical quantities and physical I / O channels are automatically calibrated, and the calibration results are updated to the data interface mapping table.

[0086] In this embodiment of the application, after generating the electrical quantity-physical I / O channel mapping table, calibration is also performed through actual hardware interaction.

[0087] Standard test excitation is a set of electrical signals with known waveform parameters and amplitudes actively applied by the edge node to the hardware input of the distribution terminal through its analog output channel. For example, it could be a sinusoidal current signal with an amplitude of 5 amps and a frequency of 50 Hz, or a sinusoidal voltage signal with an amplitude of 100 volts and an initial phase angle of 30 degrees. The hardware output response is the measured value and operational status presented on the digital output channel or communication interface of the distribution terminal after receiving the above excitation, processed by its internal current transformer, conditioning circuit, and analog-to-digital converter. Comparing the excitation and response involves comparing the known standard test excitation signal parameters with the actual response values ​​acquired and output by the terminal hardware, calculating the differences between the two. The mapping coefficient is the conversion ratio between the simulated electrical quantity and the physical I / O channel output value. For example, if 1 volt of simulated voltage corresponds to 10 volts of physical channel output, then the mapping coefficient is 0.1. Nonlinear deviation, besides the proportional relationship, refers to the error component between excitation and response caused by factors such as saturation of the internal transformer and nonlinear distortion of the conditioning circuit, which cannot be described by a single proportional coefficient. It can be recorded in the form of a deviation curve or a piecewise compensation table. Automatic calibration is the entire process where edge nodes output multiple sets of standard test excitations with different amplitudes and phase angles according to a preset procedure, automatically acquire the corresponding responses, and calculate the mapping coefficients and nonlinear deviation using algorithms such as least-squares fitting or table lookup interpolation. After updating, the data interface mapping table not only retains the original channel correspondence but also supplements the mapping coefficients and nonlinear deviation parameters obtained from calibration, enabling accurate value conversion for every electrical quantity interaction through this channel in subsequent simulations.

[0088] This calibration step is closely linked to the aforementioned S32 step of generating the mapping table, jointly improving the accuracy of signal transmission between electrical quantities and physical I / O channels. Based on the real-time capabilities of edge nodes and the requirements for transient simulation accuracy, S32 theoretically determines which electrical quantities interact through which physical channels and with what basic correspondence, thus initially establishing a signal bridge between the simulation model and the real terminal. After the initial mapping relationship is established, the simulation characteristics of the actual hardware channels, such as amplifier bias, isolator attenuation, and nonlinear distortion, may still cause deviations between the excitation and the terminal's perception. If the theoretical mapping coefficients are directly used to drive the terminal, the fault electrical quantities sensed by the terminal may exhibit amplitude shifts or waveform distortions. The calibration step, by outputting a known standard excitation and acquiring the terminal's actual response, corrects the mapping relationship from theoretical calculation to actual measurement, automatically extracting the actual mapping coefficients and nonlinear deviation parameters and updating them in the mapping table. In this way, when the simulation system injects fault electrical quantities into the terminal or reads the terminal action signal, it can perform reverse correction or forward compensation based on the calibrated parameters. This allows the terminal hardware to perceive accurate electrical information as if it were connected to the current transformer circuit of a real power distribution network, thereby improving the reliability of signal reproduction in hardware-in-the-loop simulation and providing an accurate signal link layer basis for accurately judging the terminal protection action behavior under cloud-edge collaboration.

[0089] In some embodiments of this application, it further includes: S8. During the hybrid simulation operation, monitor the computing resource utilization and cloud-edge communication round-trip latency of each edge node in real time. S9. When the resource utilization rate of any edge node is lower than the first threshold or the communication latency exceeds the second threshold, the simulation tasks corresponding to the non-critical local control simulation area undertaken by the edge node are migrated to the main simulation model in the cloud, and the area is transformed into a simplified dynamic equivalent model. S10. Update the cloud-edge collaborative simulation task sequence and the corresponding edge node data interface mapping table, reissue the updated simulation initialization parameters to each edge node, and perform clock synchronization between the cloud and edge nodes to rebalance the load between the cloud and edge nodes before continuing the hybrid simulation.

[0090] In this embodiment, S8 to S10 are further extensions of dynamically adjusting the distribution of cloud-edge tasks during the hybrid simulation process.

[0091] In S8, computational resource utilization refers to the real-time measurement of the extent to which simulation tasks consume resources such as CPU, memory, and I / O bandwidth at edge nodes during hybrid simulation operations. It can be expressed as a percentage or a normalized index. Cloud-edge communication round-trip latency refers to the bidirectional transmission time from when a query or control message is sent from the cloud to an edge node to when the cloud receives an acknowledgment or data message from the edge node. It is usually measured in milliseconds and can reflect the current network link congestion level and transmission quality.

[0092] In S9, the first threshold is a preset threshold for judging whether the computing resources of an edge node are underloaded or degraded. When the resource utilization rate is lower than this threshold, it can be understood that the remaining computing power of the edge node is insufficient to continue to stably undertake all current simulation tasks. The second threshold is a preset threshold for judging whether the cloud-edge communication quality is too poor. When the communication round-trip latency exceeds this threshold, the data exchange between the edge node and the cloud may be significantly delayed, affecting the timeliness of collaborative cooperation. Non-critical local control simulation areas refer to those areas in the edge local control simulation areas that do not undertake fast main protection outputs, have relatively low real-time requirements, or whose action consequences will not directly affect fault isolation safety, such as sections responsible for backup protection or remote backup automatic transfer logic. Simulation task migration involves transferring the transient calculation, signal driving, and data recording tasks originally assigned to a certain edge node to the cloud-based main simulation model for execution. Simplified dynamic equivalent models employ a reduced-order mathematical representation of regions migrated to the cloud. For example, a Thevenin equivalent circuit or lumped load model can replace the original detailed distributed parameter lines and multi-node topology, significantly reducing computational complexity while retaining a certain degree of dynamic response.

[0093] In S10, updating the cloud-edge collaborative simulation task sequence refers to writing the new division of labor after task migration back into the task scheduling table, ensuring that the cloud and each edge node have a consistent and clear record of their respective simulation areas. Redistributing updated simulation initialization parameters to each edge node involves redistributing the changed boundary conditions of the regions, the initial states of each node, and parameter settings due to the task re-division, ensuring that each simulation participant receives a runtime configuration matching their current division of labor. Performing clock synchronization between the cloud and edge nodes involves another round of clock calibration after task migration and parameter updates, ensuring that the cloud and edge nodes maintain a unified time base under the new division of labor. Load rebalancing refers to transferring some of the computational load from edge nodes with insufficient resources or poor communication to the cloud through the above migration and synchronization operations, thereby achieving a relative balance in the computational load and communication pressure of all simulation execution entities in the entire system.

[0094] Overall, S8 to S10 form an online adaptive adjustment mechanism of perception, decision-making, and recovery. During the continuous operation of the hybrid simulation, S8 continuously collects the resource utilization rate and cloud-edge communication round-trip latency of each edge node, expanding the previously static edge capability perception into dynamic performance monitoring that progresses with the simulation, providing real-time triggering basis for subsequent adjustment decisions. S9 judges the operating status of edge nodes based on preset thresholds and takes migration actions when the judgment is passed. It is worth noting that this does not involve moving all tasks of the entire edge node, but rather selecting a portion of the non-critical local control simulation area for migration and converting it to a simplified model. This alleviates the computational or communication pressure on the edge node while preserving its critical protection logic to execute on the local real terminal, avoiding a significant decrease in test fidelity due to large-scale task migration. After completing the migration, S10 updates the task sequence, data interface mapping table, and simulation parameters, and resynchronizes the clock. This allows the entire hybrid simulation system to quickly return to an orderly and synchronized operating state after the task re-division. Under the new division of labor, the cloud and edge nodes can continue to collaborate to advance the simulation, thereby reducing the risk of system-wide shutdown due to the performance degradation of individual edge nodes. This three-step collaboration enables the hybrid simulation to dynamically self-heal in the face of fluctuations in edge resources or deterioration in communication conditions. It allows long-term cloud-edge collaborative testing to adapt to the uncertainties of the real-world edge environment and continuously provide effective test data.

[0095] In some embodiments of this application, after updating the equivalent interface boundary conditions of the corresponding edge simulation region in the cloud-based main simulation model using the corrected equivalent input in step S63, the method further includes: S631. Collect the corrected equivalent input sequence generated in multiple consecutive synchronization cycles in the cloud, and construct the time series samples of the boundary state quantities of each edge simulation region. S632. Using a pre-trained time-series prediction model, predict the boundary state trend of the edge simulation region in the next synchronization cycle based on time-series samples, and form the predicted boundary conditions. S633. Before the next synchronization cycle arrives, the predicted boundary conditions are sent to the corresponding edge nodes in advance, so that the edge nodes can use them as the initial values ​​of the local boundary for transient solution when they have not received the next reference vector from the cloud, and complete the correction after receiving the formal reference vector.

[0096] In this embodiment of the application, after the boundary conditions of the equivalent interface are updated, a boundary condition pre-deployment mechanism based on time-series prediction is also introduced.

[0097] In S631, the synchronization cycle refers to the time interval during which a complete reference vector distribution, edge node simulation advancement, and time-stamped data feedback are completed between the cloud and edge nodes, typically corresponding to one or more global steps. The corrected equivalent input sequence refers to the array formed by arranging the equivalent input values ​​actually used to update the cloud equivalent interface boundary conditions after each compensation correction in S62 and S63 in chronological order. Boundary state quantities are electrical quantities such as voltage amplitude, phase angle, or injected power at the interface between the edge simulation region and the main cloud simulation model; these quantities determine how the edge region affects the global cloud model. A time-series sample is a set of samples formed by arranging the boundary state quantities at the same interface location in multiple consecutive synchronization cycles along a time axis, reflecting the trend of boundary state evolution as the simulation progresses.

[0098] In S632, the time-series prediction model is a mathematical model pre-trained with a large amount of historical data from cloud-edge co-simulation. It can predict the approximate trend of boundary states in the next few time steps based on the boundary state variables of the past few time steps. For example, the time-series prediction model can adopt an autoregressive model or a structure based on a long short-term memory network. Its input is a sequence of boundary state variables from the most recent several synchronization cycles, and its output is a predicted value of the boundary state for the next synchronization cycle. The predicted boundary conditions are the predicted values ​​of voltage, current, or power at the interface locations of each edge simulation region in the next synchronization cycle, output by the time-series prediction model; these are preliminary estimates of the future boundary states.

[0099] In S633, when the next reference vector from the cloud is not received, transient solution using the local boundary initial value means that at the beginning of a new synchronization cycle, if the edge node has not yet received the formally calculated reference vector from the cloud, it first uses the previously received predicted boundary conditions as temporary boundary values ​​for the external interface to drive the local simulation container to perform one or more transient solutions to maintain uninterrupted simulation progress. The formal reference vector is the interface boundary condition with higher reliability calculated by the cloud based on the global simulation state and compensation and correction algorithm within the new synchronization cycle. Completion of correction means that after receiving the formal reference vector, the edge node replaces the previously used predicted value with the formal value, correcting the state that has been calculated locally based on the predicted value, so that the state returns to a track consistent with the formal result from the cloud.

[0100] Overall, S631 to S633 introduce a feedforward path on top of the original closed-loop compensation by collecting historical correction sequences, training prediction models, and pre-issuing prediction boundaries. S631 first accumulates the equivalent input sequences of each edge simulation region after multiple rounds of synchronous correction, forming time-series samples reflecting the true changing trends of the boundary states. These samples contain both steady-state evolution patterns and implicit transient jump modes caused by faults, switching actions, etc., providing a continuously updated data foundation for the prediction model's input. S632 uses the pre-trained time-series prediction model to extrapolate the trajectory of these samples, outputting the predicted boundary conditions for the next synchronization cycle. This is equivalent to providing a trend estimate before the formal calculation in the cloud, forming a forward-looking judgment of the upcoming boundary state. S633 pre-issues this predicted value to the edge nodes when the next synchronization cycle arrives, allowing the edge nodes to start a new round of calculation without waiting for the formal reference vector to arrive in the cloud, and then perform corrections after the formal reference vector arrives. This predictive-early delivery-post-correction model can reduce the idle waiting time of edge nodes due to waiting for cloud vectors without reducing the final synchronization accuracy, allowing edge node computing resources to be used more continuously, thereby compressing the overall time consumption of the hybrid simulation and enabling cloud-edge collaborative simulation to run at a more compact pace in longer test scenarios.

[0101] In some embodiments of this application, after injecting the corresponding fault electrical analog quantity into the power distribution terminal hardware through the I / O interface and starting to collect the terminal's input and output signals in step S52, the method further includes: S521. When the edge node detects that the terminal input / output signal is inconsistent with the expected protection action logic of the local simulation within the preset response time window, record the simulation state snapshot at the current moment. S522. Keep the local simulation clock from advancing. Based on the inconsistency type, iteratively correct the injected faulty electrical analog parameters within the preset adjustment range, and re-drive the power distribution terminal hardware until the terminal action signal is consistent with the expected logic or the maximum number of iterations is reached. S523. Update the fault electrical simulation quantity according to the correction parameters at the end of the iteration, continue to advance the local simulation clock, and feed back the iteration process and correction parameters along with the time stamp data to the cloud.

[0102] In this embodiment of the application, after injecting fault electrical analog quantities into the power distribution terminal hardware and collecting input and output signals, a consistency detection and closed-loop correction step is also introduced.

[0103] In S521, the preset response time window is an allowable duration calculated from the moment the fault electrical analog quantity injection is completed. It is used to capture the protection action signals that the distribution terminal hardware should generate within this time period. Its length can be set according to the action time indicators and communication delay margin in the technical specifications of the terminal under test. Inconsistency between the terminal input / output signals and the expected protection action logic in the local simulation refers to a mismatch between the actual tripping, reclosing, or blocking switching signals acquired by the edge node and the expected action type and timing calculated in the simulation container based on the fault type, amplitude, and current terminal parameters. For example, if a tripping signal is expected to be received within 100 milliseconds after injection, but is not received within the response time window, or a reclosing signal is received, this is considered inconsistency. The simulation state snapshot is a record that packages and stores all complete state quantities in the simulation container at the current moment, including node voltages, branch currents, switch states, and internal terminal logic variables, for post-analysis of the causes of inconsistencies.

[0104] In S522, keeping the local simulation clock unadvanced means that the local simulation time of the edge node is temporarily frozen at the current moment, without performing new transient calculations. This allows for correction attempts in a time-static electrical environment, avoiding the introduction of new variables by changes in the external system state. Inconsistency types are classifications of inconsistencies, such as signal loss, incorrect signal type, and timing deviations. Different types can have different correction strategies. The preset adjustment range is the boundary set allowing changes to the faulty electrical analog parameters during the correction process. For example, the fault resistance can be adjusted between 0.1 and 100 ohms, and the fault initial phase angle can be offset by ±15 degrees, preventing correction operations from exceeding the physically feasible range. Iterative correction involves repeatedly fine-tuning the injected faulty electrical analog parameters according to a certain step size or optimization algorithm, and then driving the terminal hardware again to observe the response. The newly acquired terminal signal is re-compared with the expected logic, forming a trial-and-error convergence process. The maximum number of iterations is an upper limit set to prevent the correction process from looping infinitely. If consistency is still not satisfied after reaching this number of iterations, the correction can be terminated and the non-converged state recorded.

[0105] In S523, the correction parameter at the end of the iteration is the fault electrical analog quantity parameter that is finally determined through the above iterative correction process, enabling the terminal action signal to be consistent with the expected logic or reaching the maximum number of iterations. Updating the fault electrical analog quantity means replacing the initially preset fault injection parameter with this correction parameter, allowing subsequent simulations to continue under conditions based on more accurate excitation. Continuing to advance the local simulation clock means lifting the pause in S522 and resuming the forward flow of local simulation time. Feeding back the iterative process and correction parameters along with time-stamped data to the cloud means that the edge node packages information such as the number of iterations, the adjustment amount per step, the final correction parameter, and the convergence status in this inconsistency event as additional fields, along with the regular terminal action signal and electrical analog quantity data, and sends them to the cloud, providing more detailed local hardware interaction basis for the cloud to perform global evaluation and fault test reports.

[0106] Overall, S521 to S523 embed a consistency judgment and local correction closed loop into the unidirectional process of fault injection and signal acquisition, upgrading the hardware-in-the-loop interaction from an open-loop excitation-acquisition model to a closed-loop process with detection feedback and adaptive adjustment. S521 captures the difference between the actual terminal behavior and the simulation expectation at the first moment by comparing the expected behavior within a preset response time window, and records a state snapshot as on-site evidence. This avoids the deviation from amplifying and affecting subsequent simulations, and provides a starting reference point for correction operations. S522 pauses the local clock after confirming inconsistency, limiting the correction operation to the static electrical environment at that simulation moment, eliminating interference from continuous system evolution. Simultaneously, based on the inconsistency type, it iteratively adjusts the fault parameters and re-drives the terminal within the allowable physical range, approximating the excitation parameters that enable the real terminal to respond correctly in a closed-loop manner. S523 immediately resumes simulation progress after the correction reaches consistency or terminates, limiting the short pause caused by the correction to within the edge node, and outputs the correction parameters and the entire process record to the cloud. In this way, on the one hand, edge nodes improve the excitation matching degree of real terminals in the hardware-in-the-loop environment by adaptively correcting the fault injection amount online, reducing the inconsistency of actions caused by residual deviations in interface calibration or discrete terminal characteristics; on the other hand, the cloud obtains complete local correction process data, which helps to trace the differences in terminal behavior in the test report, evaluate the consistency of strategy execution and the adaptability of the terminal, and enhance the credibility and information integrity of the entire hybrid simulation at the signal link level.

[0107] In some embodiments of this application, after generating the test report in step S7, the method further includes: S71. Extract fault test cases that fail to meet the preset collaborative performance indicators from the test report. Analyze the key control parameters that cause performance deviations for the fault location accuracy, isolation coordination, and power restoration time indicators in the extracted test cases. S72. Based on the control logic of the automation strategy of the feeder under test, an adjustment vector for key control parameters is generated, and the control parameters of the cloud-based centralized decision-making simulation area and the edge-based local control simulation area are modified according to the adjustment vector. S73. Update the cloud-edge collaborative simulation task sequence and fault scenario timestamp sequence according to the modified control parameters, re-trigger the hybrid simulation, and perform regression verification on the adjusted strategy until the collaborative performance indicators of all fault test cases meet the requirements.

[0108] In this embodiment of the application, after generating the test report, a closed-loop process from evaluation feedback to automatic optimization of strategy parameters and regression verification is also introduced.

[0109] In S71, preset coordination performance indicators are pre-defined qualification criteria for evaluation dimensions such as fault location accuracy, isolation coordination, and power restoration time. Examples include a fault location accuracy rate not lower than a certain percentage, a coordination time difference between adjacent switch protections falling within a certain millisecond range, and a net power restoration time shorter than a certain time limit. Fault test cases that fail to meet preset coordination performance indicators refer to test scenarios where one or more of the above evaluation results failed to meet the preset qualification criteria in the previous round of hybrid simulation. Performance deviation refers to the difference between the measured indicator value and the preset indicator, which can be described in terms of time deviation, accuracy difference, or specific forms of coordination mismatch. Key control parameters are parameters that significantly affect the decision-making behavior and timing performance of the feeder automation strategy under test in the fault scenario, such as overcurrent protection start setting, delay setting, reclosing charging time, power transfer waiting time limit, and critical conditions for cloud-edge task partitioning. Analysis of key control parameters that cause performance deviations can be implemented using sensitivity analysis methods. This involves sequentially perturbing each candidate control parameter and observing the change in performance indicators, identifying parameters highly correlated with these changes as key control parameters.

[0110] In S72, the adjustment vector is a set of data consisting of the modification direction and amount of key control parameters, indicating whether each key parameter should be increased or decreased, and the specific value or percentage of the adjustment. The generation of the adjustment vector can be determined based on the direction and magnitude of performance deviations. For example, if the isolation coordination deviation manifests as excessive edge local control delay, the adjustment amount can be generated in the direction of shortening the delay setting value. Modifying the control parameters in the cloud-based centralized decision-making simulation area refers to changing the global coordination logic parameters deployed in the cloud, such as adjusting the priority weight of the transfer path selection or the time coordination value of cross-regional backup protection. Modifying the control parameters in the edge local control simulation area refers to changing the local protection settings and reclosing logic parameters deployed at the edge nodes. Control parameters can include protection parameters such as starting current, operating time, and reset coefficient, as well as logic parameters such as coordination level difference, blocking conditions, and self-healing recovery delay. For example, the adjustment vector can be formed as a mapping table of parameter names and adjustment values, which can be directly read and written into the parameter configuration file of the corresponding simulation area by an automated script.

[0111] In S73, updating the cloud-edge collaborative simulation task sequence refers to checking whether the cloud-edge task division needs adjustment due to changes in control parameters. For example, if the action delay of a certain edge region is significantly shortened, it may change from a non-critical region to a critical region. Or, if parameter adjustments affect the execution entity to which the coordination logic belongs, the relevant entries in the task sequence need to be revised synchronously. Updating the fault scenario timestamp sequence means that when re-triggering the simulation, the same fault occurrence time and fault type as the original test are used, but the fault duration or fault clearing method may change due to parameter adjustments, and these are recalculated and set. Regression verification refers to re-executing the hybrid simulation on the feeder automation strategy after parameter adjustments using the same or partially the same fault test cases as the previous round of testing to confirm whether the adjustment measures have effectively eliminated the previous performance deviations and have not introduced new coordination problems. Meeting the coordination performance requirements for all fault test cases means that after one or more rounds of regression verification, the fault location accuracy, isolation coordination, and power restoration time in each fault scenario all meet the qualified criteria of the preset coordination performance indicators.

[0112] Overall, S71 to S73 constitute a complete closed loop from identifying deficiencies in test evaluation, to targeted adjustment of strategy parameters, and finally to regression testing to confirm the effect. S71 automatically filters out faulty test cases that have not yet met the standards by comparing the measured performance in the test report with preset collaborative performance indicators, and identifies key control parameters that significantly affect the test results based on indicator deviations. This step transforms test evaluation from simple data recording into diagnostically meaningful deviation localization, providing a clear target for subsequent adjustments. S72 then utilizes the control logic structure of the strategy under test itself to transform performance deviations into specific adjustment vectors for key control parameters. It modifies the parameter configurations of the centralized decision-making part in the cloud and the local control part at the edge, according to the cloud-edge division of labor. This ensures that the adjustment operation covers both global coordination and local response levels, avoiding new imbalances in cloud-edge collaboration caused by adjusting only one aspect. S73 refreshes the task sequence and fault sequence after parameter modification, re-triggering hybrid simulation for regression verification with the updated configuration. This process can be iterated until all test cases meet the standards. In this way, the whole method is no longer limited to a single test and evaluation, but integrates simulation testing and strategy optimization. At the operational level, it forms an automatic cycle of test-diagnosis-tuning-re-verification, which helps to reduce the workload of repeated manual parameter tuning and makes the performance of the strategy under cloud-edge collaborative conditions more mature and stable after multiple rounds of verification.

[0113] A second aspect of this application provides a cloud-edge collaborative automated hybrid simulation test platform for feeders, including a cloud server and multiple edge simulation nodes; The cloud server is equipped with a main simulation module and a collaborative management module, while the edge simulation nodes are equipped with a hardware-in-the-loop simulation module and an interface adaptation module. The main simulation module is used to build and run the cloud-based feeder digital twin main simulation model; The collaborative management module is used to parse the control logic partition of the automation strategy of the feeder under test, generate the cloud-edge collaborative simulation task sequence, dynamically determine the simulation step size, data interface mapping table and simulation clock reference of each edge simulation node, and execute the collaborative control and evaluation functions in the above method. The hardware-in-the-loop simulation module is used to run the hardware-in-the-loop simulation container of the power distribution terminal on the corresponding edge simulation node, drive the power distribution terminal hardware and acquire signals; The interface adapter module is used to realize physical I / O connection and signal mapping with the power distribution terminal hardware.

[0114] A third aspect of this application provides an electronic device, including: one or more processors, one or more input devices, one or more output devices, and one or more memories. The processors, input devices, output devices, and memories communicate with each other via a communication bus. The memories store computer programs, including program instructions. The processors execute the program instructions stored in the memories. The processors are configured to invoke the program instructions to execute the aforementioned cloud-edge collaborative feeder automation hybrid simulation testing method.

[0115] It should be understood that, in the embodiments of this application, the processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0116] Input devices may include touchpads, fingerprint sensors (for collecting the user's fingerprint information and fingerprint orientation information), microphones, etc., while output devices may include displays (LCDs, etc.), speakers, etc.

[0117] The memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store information about the device type.

[0118] In specific implementations, the processor, input device, and output device described in the embodiments of this application can execute the implementation methods described in any embodiment of the cloud-edge collaborative feeder automation hybrid simulation test method provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0119] In another embodiment of this application, an electronic device is provided. The electronic device stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the cloud-edge collaborative feeder automation hybrid simulation testing method described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in an electronic device, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0120] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0121] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0122] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0123] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0125] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0126] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A hybrid simulation testing method for feeder automation based on cloud-edge collaboration, characterized in that, Includes the following steps: S1. Based on the automation strategy of the feeder under test and the topology of the distribution line, a cloud-edge hybrid simulation model is constructed. The hybrid simulation model includes a cloud-edge feeder digital twin master simulation model and a distribution terminal hardware-in-the-loop simulation container deployed on each edge node. S2. Analyze the control logic of the automation strategy of the feeder under test, divide the feeder topology into a cloud-based centralized decision-making simulation area and at least one edge-based local control simulation area, and generate a cloud-edge collaborative simulation task sequence. S3. For each edge local control simulation area, obtain the real-time computing resources and physical I / O interface status of the corresponding edge node, and dynamically determine the simulation step size, data interface mapping table and simulation clock reference of the edge node. S4. After triggering the hybrid simulation, according to the cloud-edge collaborative simulation task sequence, the simulation initialization parameters and the fault scenario timestamp sequence containing the fault occurrence time and fault type are sent to each edge node, and the global simulation clock is started. S5. Control each edge node to drive the power distribution terminal hardware to run in the hardware-in-the-loop simulation container according to the received simulation initialization parameters and fault scenario timestamp sequence, collect terminal protection action signals and electrical analog quantities in real time, and receive time-stamped data fed back by each edge node after adding edge timestamps. S6. Use the received time-stamped feedback data to perform spatiotemporal alignment with the simulation state of the main simulation model in the cloud at the corresponding time, identify data gaps and timing misalignments, use a compensation and correction algorithm based on reference vector interpolation to adjust the synchronization of the cloud-edge simulation state, and control the global simulation clock to advance according to the slowest edge node. S7. When the preset simulation end condition is reached, stop the hybrid simulation. Based on the full process state variables recorded in the cloud and the terminal action event sequence fed back by each edge node, evaluate the fault location accuracy, isolation coordination and power restoration time of the feeder automation strategy under cloud-edge collaboration, and generate a test report.

2. The method according to claim 1, characterized in that, The step S3, which dynamically determines the simulation step size, data interface mapping table, and simulation clock reference for the edge node, specifically includes: S31. Obtain the CPU utilization, memory availability, and I / O interrupt response latency of the edge node, and calculate the real-time simulation capability index of the node. S32. Based on the real-time simulation capability index and the electrical transient simulation accuracy required for the edge local control simulation area, set the minimum simulation step size and generate the electrical quantity-physical I / O channel mapping table of the power distribution terminal hardware terminals under the simulation step size. S33. Based on the cloud-based global simulation clock, synchronize the clock of the edge node and generate the local simulation propagation clock reference of the edge node on the global clock scale according to the set simulation step size.

3. The method according to claim 1, characterized in that, Step S6 employs a compensation and correction algorithm based on reference vector interpolation to adjust the cloud-edge simulation state synchronization, specifically including: S61. The node voltage amplitude and phase angle at the current moment of the cloud-based main simulation model are used as reference vectors and compared with the node electrical quantities with edge time scales fed back by the edge nodes at the same moment to form an observation vector, and the timing misalignment is detected. S62. If the timing misalignment exceeds a set threshold, an electrical quantity difference vector between adjacent step sizes of the edge node is constructed in the cloud, and the observation vector is linearly interpolated to generate a corrected cloud equivalent input. S63. Update the equivalent interface boundary conditions of the corresponding edge simulation region in the cloud master simulation model using the corrected equivalent input, drive the cloud simulation state correction, and send the corrected reference vector to the corresponding edge node for the edge node to use for the next long initial value correction.

4. The method according to claim 1, characterized in that, In step S4, a fault scenario timestamp sequence containing the fault occurrence time and fault type is sent to each edge node, including: S41. Based on the fault test cases of the automated strategy of the feeder under test in the cloud, generate a sequence of fault events marked with a unified absolute timestamp for the entire system. Each fault event includes the fault type, fault location identifier and duration. S42. Based on the division of the cloud-edge collaborative simulation task sequence, the fault event sequence is split into local fault timestamp sub-sequences that only involve the corresponding edge local control simulation area; S43. Along with the simulation initialization parameters, the local fault timestamp subsequence is sent to the corresponding edge nodes, while the global fault event sequence containing all edge regions is retained in the cloud-based main simulation model for the main simulation to advance.

5. The method according to claim 1, characterized in that, In step S5, controlling each edge node to drive the power distribution terminal hardware within the hardware-in-the-loop simulation container, acquiring terminal protection action signals and electrical analog quantities in real time, and receiving time-stamped data fed back by each edge node after adding edge time stamps, specifically includes: S51. Create a real-time thread at the edge node that matches the simulation step size, and iteratively solve the transient equations of the power system in the hardware-in-the-loop simulation container under the drive of the local simulation advancement clock reference. S52. When the local simulation clock advances to the fault occurrence time in the received fault scenario timestamp sequence, the corresponding fault electrical analog quantity is injected into the power distribution terminal hardware through the I / O interface, and the terminal's input and output signals are collected. S53. The edge node combines the collected terminal action signal with the local simulated electrical quantity at the current time, and adds a double time stamp containing the local simulated clock time and the actual physical time to form the time-stamped feedback data.

6. The method according to claim 3, characterized in that, Step S6, which controls the global simulation clock to advance according to the slowest edge node, specifically includes: S64: Set a global simulation step counter in the cloud and collect the latest local simulation clock time from all edge nodes at each global step time; S65. Identify the minimum advance time among all local simulation clock times, adjust the global simulation clock to the minimum advance time, and determine whether there is a continuous delay in the lagging edge nodes. S66. If the delay of multiple consecutive steps of an edge node exceeds the preset tolerance threshold, the cloud-based main simulation model performs model reduction processing on the simulation area corresponding to the edge node, converts the power distribution lines of the area into a lumped impedance model, and increases the calculation step size to accelerate its synchronization.

7. The method according to claim 1, characterized in that, Step S7, which evaluates the fault location accuracy, isolation coordination, and power restoration time of the feeder automation strategy under cloud-edge collaboration, specifically includes: S71. Extract the overcurrent and undervoltage signals of each switch before and after the fault moment from the cloud-based full-process state variables, and perform causal correlation with the terminal action event sequence fed back by each edge node to calculate the fault location accuracy. S72. Analyze the time difference between the protection actions and reclosing coordination of adjacent switches, combine the recorded cloud-edge interaction delay, evaluate the coordination of isolation actions, and distinguish the contribution ratio of cloud decision delay and edge local control delay. S73. Calculate the time from the occurrence of the fault to the restoration of power supply to the non-faulty section, and subtract the known inherent delay of I / O response and communication delay in the hybrid simulation to obtain the net power restoration time index.

8. The method according to claim 2, characterized in that, After generating the electrical quantity-physical I / O channel mapping table in step S32, the method further includes: The control edge node outputs standard test excitation to the power distribution terminal hardware input terminal and collects the hardware output response. By comparing the excitation and response, the mapping coefficient and nonlinear deviation between electrical quantities and physical I / O channels are automatically calibrated, and the calibration results are updated to the data interface mapping table.

9. The method according to claim 1, characterized in that, The method further includes: S8. During the hybrid simulation operation, monitor the computing resource utilization and cloud-edge communication round-trip latency of each edge node in real time. S9. When the resource utilization rate of any edge node is lower than the first threshold or the communication latency exceeds the second threshold, the simulation tasks corresponding to the non-critical local control simulation area undertaken by the edge node are migrated to the main simulation model in the cloud, and the area is transformed into a simplified dynamic equivalent model. S10. Update the cloud-edge collaborative simulation task sequence and the corresponding edge node data interface mapping table, reissue the updated simulation initialization parameters to each edge node, and perform clock synchronization between the cloud and edge nodes to rebalance the load between the cloud and edge nodes before continuing the hybrid simulation.

10. A cloud-edge collaborative feeder automation hybrid simulation test platform, characterized in that, Includes cloud servers and multiple edge simulation nodes; The cloud server is equipped with a main simulation module and a collaborative management module, and the edge simulation node is equipped with a hardware-in-the-loop simulation module and an interface adaptation module. The main simulation module is used to build and run the cloud-based feeder digital twin main simulation model; The collaborative management module is used to parse the control logic partition of the automation strategy of the feeder under test, generate a cloud-edge collaborative simulation task sequence, dynamically determine the simulation step size, data interface mapping table and simulation clock reference of each edge simulation node, and execute the collaborative control and evaluation functions in any one of claims 1 to 9. The hardware-in-the-loop simulation module is used to run the hardware-in-the-loop simulation container of the power distribution terminal on the corresponding edge simulation node, drive the power distribution terminal hardware and collect signals. The interface adapter module is used to realize physical I / O connection and signal mapping with the power distribution terminal hardware.

Images