Adaptive setting and action verification method and system for spacer protection device

Abstract

The application provides an adaptive setting and action verification method and system for interval layer protection devices, relates to the technical field of power system automation, and acquires historical fault current and voltage waveform data of a power system, extracts current characteristic waveforms and voltage characteristic waveforms corresponding to a target fault type from the historical fault current and voltage waveform data, and processes the current characteristic waveforms and the voltage characteristic waveforms to obtain target characteristic waveform data. In combination with real-time operation parameters of a power grid, a multi-agent collaborative strategy of an interval layer protection device of a transformer substation is generated to determine a target candidate action threshold and a target candidate delay parameter. In combination with a virtual transformer substation system which is identical to an actual transformer substation structure, action response records of the interval layer protection device of the transformer substation under multiple fault scenarios are simulated. According to a comparison result of the action response records and a preset response standard set, an action verification conclusion of an adaptive setting process is generated, real-time closed-loop verification after parameter adjustment is realized, and the adaptive capability of the protection system under a variable operation environment is improved.

Description

Technical Field

[0001] This application relates to the field of power system automation technology, and in particular to an adaptive setting and action verification method and system for bay layer protection devices. Background Technology

[0002] With the development of new power systems, the integration of numerous distributed power sources has led to increasingly complex and variable power grid operation modes. Substation bay layer protection devices face challenges such as diversified fault characteristics and uncertain power flow directions. In actual operation, protection devices need to maintain high sensitivity and selectivity under different operating conditions. Especially under power grid conditions, traditional fixed setting methods are difficult to adapt to dynamically changing system conditions and are prone to false tripping or failure to trip.

[0003] Current research attempts to collect synchronous phasor data from multiple substations using a wide-area measurement system, combining it with an offline setting database and rule engine to match pre-set value combinations corresponding to the current operating mode online, and then distribute these values ​​to the bay-level protection devices to achieve setting switching. However, existing solutions rely on pre-set value tables, which cannot cope with complex operating conditions or new fault modes that are not covered, and lack the ability to autonomously generate optimal parameters, resulting in limited adaptability under extreme or atypical operating conditions. The offline simulation verification method used has high computational overhead and is time-consuming, making it difficult to achieve real-time closed-loop verification synchronized with protection settings, and failing to promptly detect coordination errors or action delays that may be caused by parameter adjustments. Summary of the Invention

[0004] The purpose of this application is to provide an adaptive setting and action verification method and system for bay layer protection devices, in order to solve the problems in the existing technology, such as the reliance on preset value tables, which makes it difficult to adapt to complex working conditions and lacks autonomous optimization capabilities; offline simulation is time-consuming, making it difficult to achieve real-time closed-loop verification and unable to detect action coordination in a timely manner.

[0005] To address the aforementioned technical problems, in a first aspect, this application provides an adaptive setting and operation verification method for bay layer protection devices, comprising: Collect historical fault current and voltage waveform data of the power system, and extract the current characteristic waveform and voltage characteristic waveform corresponding to the target fault type from the historical fault current and voltage waveform data; By configuring edge computing nodes in the substation bay layer, the current characteristic waveform and voltage characteristic waveform are preprocessed and compressed to obtain target characteristic waveform data. Based on the real-time operating parameters of the power grid and the target characteristic waveform data, a multi-agent collaborative strategy for the substation bay layer protection device is generated through a deep reinforcement learning network. The real-time operating parameters include power output value, power grid flow direction and bus voltage level. The target candidate action threshold and target candidate delay parameter of the substation bay layer protection device are determined by the multi-agent cooperative strategy. Using digital twins, a virtual substation system with the same structure as the actual substation is constructed. Combining the target candidate action threshold and the target candidate delay parameter, the action response records of the substation bay layer protection device are simulated under various fault scenarios. Based on the comparison results of the action response records with the preset response standard set, the action verification conclusion of the adaptive setting process is generated.

[0006] Optionally, based on the real-time operating parameters of the power grid and the target characteristic waveform data, a multi-agent cooperative strategy for the substation bay layer protection device is generated through a deep reinforcement learning network, including: The substation bay layer is divided into multiple protection zones, and a corresponding intelligent agent is configured for each protection zone. Define a communication protocol between the agents, establish an information interaction channel between agents in adjacent protection zones based on the communication protocol, and transmit the operating status parameters of the substation bay layer protection devices associated with each agent through the information interaction channel to form a state sharing mechanism between agents. Based on the fault characteristics in the target characteristic waveform data, and combined with the bus voltage level in the real-time operating parameters, the action constraint range is determined. Based on the real-time operating parameters and the action constraint range, the action threshold adjustment direction and delay parameter scaling ratio corresponding to each agent are calculated through a deep reinforcement learning network to generate the preliminary strategy corresponding to each agent. Based on the state sharing mechanism, each preliminary strategy is transmitted through the information interaction channel to identify conflicting parameter adjustment contents in each preliminary strategy. Through the interaction and cooperation between agents, the conflicting parameter adjustment contents are corrected to obtain the agent strategy. Based on the electrical connection relationship of each protected area, the agent strategies are correlated and integrated to form a multi-agent collaborative strategy.

[0007] Optionally, based on the real-time operating parameters and the action constraint range, a deep reinforcement learning network is used to calculate the action threshold adjustment direction and delay parameter scaling ratio for each agent to generate a preliminary strategy for each agent, including: In the policy calculation module configured in each agent, the signal duration of fault characteristics is converted into time quantization value, the amplitude change range is converted into amplitude quantization value, and the power output value, power flow direction and bus voltage level in real-time operating parameters are converted into power output quantization value, enumeration value and voltage standard value respectively to form a state quantization set. Based on the state quantization set, the adjustment action options that the agent can perform are set; Based on the adjustable range of the action threshold within the action constraint range, the maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio are set for the forward adjustment of the action threshold, the reverse adjustment of the action threshold, the forward scaling of the delay parameter, and the reverse scaling of the delay parameter, respectively. Using deep reinforcement learning, the state quantization set, maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio are iteratively calculated to generate the corresponding action value. Select the action threshold adjustment direction and delay parameter scaling ratio corresponding to the action with the highest action value, and generate a preliminary strategy for the corresponding agent based on the action threshold adjustment direction and delay parameter scaling ratio.

[0008] Secondly, this application provides an adaptive setting and operation verification system for bay layer protection devices, including: The acquisition module is used to acquire historical fault current and voltage waveform data of the power system, and extract the current characteristic waveform and voltage characteristic waveform corresponding to the target fault type from the historical fault current and voltage waveform data; The processing module is used to preprocess and compress the current characteristic waveform and voltage characteristic waveform through edge computing nodes configured in the substation bay layer to obtain target characteristic waveform data. The generation module is used to generate a multi-agent cooperative strategy for the substation bay layer protection device through a deep reinforcement learning network based on the real-time operating parameters of the power grid and the target feature waveform data. The real-time operating parameters include power output value, power grid flow direction and bus voltage level. The determination module is used to determine the target candidate action threshold and target candidate delay parameters of the substation bay layer protection device through the multi-agent collaborative strategy; The module is used to construct a virtual substation system with the same structure as the actual substation using digital twins. It combines the target candidate action threshold and the target candidate delay parameter to simulate the action response records of the substation bay layer protection device under various fault scenarios. Based on the comparison results of the action response records with the preset response standard set, it generates the action verification conclusion of the adaptive setting process.

[0009] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the adaptive tuning and action verification method for a spacer protection device as described in the first aspect above.

[0010] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the adaptive tuning and action verification method for a spacer protection device as described in the first aspect above.

[0011] The adaptive setting and action verification method for bay-level protection devices provided in this application collects historical fault current and voltage waveform data of the power system and extracts characteristic waveforms corresponding to the target fault type, providing accurate fault basis for subsequent parameter setting. It preprocesses and compresses the characteristic waveforms through edge computing nodes in the substation bay level, ensuring data quality while reducing data volume and improving subsequent computational efficiency. Combining real-time power grid operating parameters with target characteristic waveform data, it generates a multi-agent collaborative strategy through a deep reinforcement learning network, overcoming the limitations of traditional fixed-value setting and achieving autonomous adaptation to dynamic power grid conditions. Based on this collaborative strategy, it determines the target candidate action threshold and target candidate delay parameters of the protection device, ensuring the targeted and reasonable adjustment of parameters. Using a digital twin, it constructs a virtual system consistent with the actual substation, simulating various fault scenarios and verifying parameter action responses, achieving a real-time closed loop of setting and verification. This effectively avoids the risk of false or failed operation due to improper parameters, comprehensively improving the adaptive capability and operational reliability of the protection system.

[0012] Furthermore, by acquiring the physical structural dimensions, bay layer protection device installation locations, and electrical connection line parameters of the actual substation, a virtual spatial layout, virtual protection device deployment locations, and virtual connection lines are set to form a virtual substation system. Virtual identifiers are assigned to the virtual protection devices, and a correspondence table between them and the actual devices is established. Based on this correspondence table, target candidate action thresholds and delay parameters are matched to the corresponding virtual identifiers to form a set of virtual protection devices. Multiple sets of simulated fault scenario parameters with scenario numbers are generated according to the target fault type. Virtual line faults are simulated in the virtual substation system according to the scenario numbers, and action response records are recorded. Based on relay protection... The protection system uses four criteria and scenario numbers to determine a set of preset response standards, including standard action time intervals, types, and parameter value ranges. Action response records under the same scenario number are compared with the preset response standard set to generate comparison results and action verification conclusions for the adaptive setting process. This achieves dynamic modeling consistent with the actual structure and real-time closed-loop verification synchronized with protection setting. It can promptly detect misalignment or action delays caused by parameter adjustments. Through dynamic modeling and rapid deduction mechanisms, the adaptability, action reliability, continuous optimality of protection performance, and immediate reliability of action logic of the protection system in rapidly changing power grid environments are improved. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating the adaptive tuning and action verification method for a spacer protection device provided in this application embodiment; Figure 2 A schematic diagram illustrating a specific implementation of the adaptive tuning and action verification method for a spacer protection device provided in this application. Figure 3 This is a schematic diagram of the adaptive tuning and action verification system for a spacer protection device provided in an embodiment of this application. Detailed Implementation

[0015] To address the issues of poor adaptability and insufficient reliability of traditional setting values ​​in substation protection devices operating under power grids due to the variable operating modes, existing technologies rely on preset value tables and offline simulations, which are insufficient for handling unmodeled operating conditions and lack real-time closed-loop verification capabilities. Therefore, this application acquires historical fault waveforms and processes them through edge nodes to obtain typical features. Combined with real-time operating parameters, a deep reinforcement learning network is used to dynamically generate multi-device collaborative adjustment strategies, achieving autonomous optimization of protection thresholds and delays. Simultaneously, a virtual substation system synchronized with the physical system is constructed to perform rapid multi-scenario simulations of the adjusted parameters, enabling immediate verification of the action response.

[0016] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] The core of this application is to provide an adaptive setting and operation verification method for spacer protection devices. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes: Step 101: Collect historical fault current and voltage waveform data of the power system, and extract the current characteristic waveform and voltage characteristic waveform corresponding to the target fault type from the historical fault current and voltage waveform data.

[0018] In this step, historical fault current and voltage waveform data refer to the waveform data of current and voltage changes over time during the fault period recorded by monitoring equipment during the past operation of the power system when a fault occurred, reflecting the changes in electrical quantities of the power system at the time of the fault. The target fault type refers to the specific type of power system fault targeted by this technical solution, such as three-phase short-circuit faults, single-phase grounding faults, and distributed generation grid connection interface faults, which are pre-defined based on common fault modes and protection requirements of the power system. Current characteristic waveforms and voltage characteristic waveforms refer to waveform segments selected from historical fault current and voltage waveform data that reflect the core electrical characteristics of the target fault type.

[0019] In this embodiment of the application, historical fault current waveform data and historical fault voltage waveform data recorded during past faults are collected through existing historical databases of the power system or long-term operation monitoring equipment. Subsequently, based on the predetermined target fault type, such as line short-circuit fault or equipment grounding fault, the typical performance characteristics of the target fault type in the current and voltage dimensions are analyzed, such as the pattern of sudden current increase and voltage drop during short-circuit faults. Waveform segments that can reflect the core characteristics of the target fault type are selected and extracted from the collected historical fault current and voltage waveform data, respectively forming current characteristic waveforms and voltage characteristic waveforms corresponding to the target fault type.

[0020] Step 102: By configuring edge computing nodes in the substation bay layer, preprocess and compress the current characteristic waveform and voltage characteristic waveform to obtain the target characteristic waveform data.

[0021] In this step, the target characteristic waveform data refers to the data set obtained after preprocessing and compressing the current characteristic waveform and voltage characteristic waveform through the edge computing node of the substation bay layer.

[0022] In this embodiment of the application, step 102 specifically includes the following steps: Step 201: Using the time of the fault occurrence as the time reference point, by configuring the edge computing node in the substation bay layer, extract the preset waveform segments from the current characteristic waveform and voltage characteristic waveform, divide the preset waveform segments into multiple initial waveform segments, and add a timestamp identifier to each initial waveform segment to obtain the intermediate waveform segment.

[0023] In this step, the fault occurrence time refers to the specific time point at which the fault begins to appear in the power system. It is obtained based on the fault trigger signal recorded by the power system monitoring equipment, reflecting the time node of fault initiation and serving as a reference for subsequent waveform extraction and time stamping. The time reference point refers to a time reference point set with the fault occurrence time as the origin. It is used to unify the time coordinates of each segment in the current and voltage characteristic waveforms, ensuring time consistency in subsequent waveform segmentation and integration processes, and avoiding feature extraction deviations caused by inconsistent time coordinates. The preset waveform segment refers to a current or voltage characteristic waveform segment with a pre-set duration centered on the time reference point, based on the typical duration of the fault characteristics. The initial waveform segment refers to a small waveform unit obtained by dividing the preset waveform segment according to fixed time intervals. This division is based on the need for refined waveform data processing and is used for subsequent calculation of signal amplitude change rate, achieving segmented capture of fault characteristics. The timestamp identifier refers to a timestamp added to each initial waveform segment and associated with the time reference point, generated based on the time reference point. The intermediate waveform segment refers to a set of initial waveform segments with timestamp identifiers, including the complete time series and segmented waveform information of the preset waveform segment.

[0024] In this embodiment, the specific time point when a power system fault occurs is first determined as the fault occurrence time, and this fault occurrence time is set as the time reference point. By configuring edge computing nodes in the substation bay layer, waveform segments of a preset duration are extracted from the current characteristic waveform and voltage characteristic waveform with the time reference point as the center. These waveform segments are called preset waveform segments. Subsequently, the preset waveform segments are divided into multiple continuous small waveform segments according to fixed time intervals. Each small waveform segment is an initial waveform segment. A timestamp identifier associated with the time reference point is added to each initial waveform segment. Finally, all initial waveform segments with timestamp identifiers are integrated to obtain intermediate waveform segments.

[0025] Step 202: Compare the rate of change of signal amplitude of adjacent intermediate waveform segments to identify abnormal waveform segments whose rate of change of signal amplitude exceeds a preset range.

[0026] In this step, the signal amplitude change rate refers to the ratio of the difference in signal amplitude between two adjacent intermediate waveform segments to the time interval between the two segments, calculated based on the amplitude data and timestamp of the intermediate waveform segments. The preset range refers to a reasonable range of signal amplitude change rates pre-defined according to the signal amplitude variation patterns during normal power system operation, obtained based on historical normal operation data, and serves as a criterion for identifying abnormal waveform segments. Abnormal waveform segments refer to intermediate waveform segments whose signal amplitude change rate exceeds the preset range, reflecting a fault-induced amplitude abrupt change in the waveform, and are key areas for subsequent extraction of core fault features.

[0027] In this embodiment, the signal amplitude difference between two adjacent intermediate waveform segments is calculated, and then the amplitude difference is divided by the time interval between the two segments to obtain the signal amplitude change rate of the adjacent intermediate waveform segments; a preset range is retrieved based on the normal fault characteristics of the power system; the calculated signal amplitude change rate is compared with the preset range one by one, and intermediate waveform segments whose signal amplitude change rate exceeds the preset range are identified as abnormal waveform segments.

[0028] Step 203: Using the peak point and valley point of the signal amplitude in the abnormal waveform segment as the boundary, extract the target waveform segment from the abnormal waveform segment.

[0029] In this step, the signal amplitude peak point refers to the specific location where the signal amplitude reaches its highest value in the abnormal waveform segment. It is identified based on the amplitude data scan of the abnormal waveform segment and reflects the location of the maximum intensity of the fault characteristic signal. The signal amplitude trough point refers to the specific location where the signal amplitude reaches its lowest value in the abnormal waveform segment. It is identified based on the amplitude data scan of the abnormal waveform segment and reflects the location of the minimum intensity of the fault characteristic signal. The target waveform segment refers to the waveform portion that includes the complete fault fluctuation cycle, extracted from the abnormal waveform segment with the signal amplitude peak point and trough point as boundaries.

[0030] In this embodiment, the waveform analysis function of the edge computing node is used to identify the points where the signal amplitude reaches the highest value and the points where the signal amplitude reaches the lowest value within the abnormal waveform segment; using the peak point and valley point of the signal amplitude as the interception boundary, a target waveform segment including the complete fluctuation cycle from the valley point to the peak point and then to the next valley point is intercepted from the abnormal waveform segment.

[0031] Step 204: Perform signal denoising processing on the target waveform segment to remove high-frequency interference signals and obtain a denoised waveform segment.

[0032] In this step, high-frequency interference signals refer to electromagnetic interference signals in the target waveform segment whose frequency is much higher than that of the fault characteristic signals. These signals are generated by electromagnetic interference sources in the power system operating environment and can mask the core fault characteristics. They are irrelevant signals that need to be removed in subsequent denoising processes. The denoised waveform segment refers to the waveform portion obtained after filtering the target waveform segment. It is obtained by removing high-frequency interference signals and retaining only the low-frequency effective signals related to the fault characteristics, thus eliminating the influence of interference on fault feature extraction.

[0033] In this embodiment of the application, filtering is used to denoise the target waveform segment. The high-frequency interference signal refers to the electromagnetic interference signal in the target waveform segment whose frequency is much higher than the frequency of the fault characteristic signal. The filtering process retains the low-frequency signal related to the fault characteristics and removes the high-frequency interference signal, finally obtaining a denoised waveform segment without interference and containing only the core fault signal.

[0034] Step 205: Perform multi-scale decomposition on the denoised waveform segment to obtain sub-waveform components in different frequency ranges, wherein each sub-waveform component corresponds to a frequency label; In this step, the frequency range refers to the different frequency ranges into which the denoised waveform segment is divided using multi-scale decomposition technology, based on the typical frequency distribution of the fault characteristic signal. The sub-waveform component refers to the waveform portion of the denoised waveform segment within each frequency range. The frequency label refers to the identifier assigned to each sub-waveform component, corresponding to its respective frequency range.

[0035] In this embodiment, the denoised waveform segment is processed by multi-scale decomposition technology, which decomposes the denoised waveform segment into multiple waveform parts with different frequency ranges. Each frequency range is a frequency interval. The waveform part corresponding to each frequency interval is a sub-waveform component. At the same time, each sub-waveform component is labeled with a frequency label corresponding to its frequency interval, such as a 0-20Hz sub-component and a 20-50Hz sub-component.

[0036] Step 206: Based on the frequency label, calculate the signal duration and amplitude variation range of the key sub-waveform components; In this step, fault characteristics refer to signal features that reflect the core electrical attributes of the target fault type, summarized based on the typical electrical performance of the target fault type. Key sub-waveform components refer to the sub-waveform components within their respective frequency ranges that correspond to the fault characteristics; these components contain the most concentrated signal information about the fault characteristics. Signal duration refers to the total time from the appearance of the key sub-waveform component to its complete disappearance. Amplitude variation range refers to the numerical range between the highest and lowest amplitude values ​​of the key sub-waveform component.

[0037] In this embodiment of the application, based on the frequency label, the sub-waveform components corresponding to the frequency range and the fault characteristics are selected. For example, for short circuit faults, the sub-waveform components of 0-20Hz are selected. These sub-waveform components are the key sub-waveform components. Then, the total time length from the appearance to the disappearance of the key sub-waveform components is counted. This length is the signal duration. At the same time, the amplitude variation range between the highest and lowest amplitudes of the key sub-waveform components is counted.

[0038] Step 207: Based on the timestamp identifier, integrate the key sub-waveform components, the signal duration, and the amplitude variation range to generate target feature waveform data; In this embodiment of the application, the key sub-waveform components, signal duration, and amplitude variation range are integrated one-to-one according to the order of the timestamps. For example, the key sub-waveform component corresponding to the timestamp 00:00:00.01 is associated with its corresponding signal duration of 0.2 seconds and amplitude variation range of 5-18A to form target feature waveform data that is concise in data volume and retains the core characteristics of the fault.

[0039] The embodiments of this application eliminate the masking of fault characteristics by high-frequency interference, while simplifying the amount of data and retaining the core fault information. This solves the problem in the background technology that traditional solutions have large deviations in subsequent parameter tuning due to interference and large redundancy in the data, and helps to generate protection strategies that are adapted to complex working conditions of multi-source grid connection.

[0040] Step 103: Based on the real-time operating parameters of the power grid and the target characteristic waveform data, generate a multi-agent collaborative strategy for the substation bay layer protection device through a deep reinforcement learning network. The real-time operating parameters include power output value, power flow direction, and bus voltage level.

[0041] In this step, the real-time operating parameters of the power grid refer to a set of real-time data including power output, power flow direction, and bus voltage level, reflecting the actual operating status of the power system under the current power grid.

[0042] In the embodiments of this application, such as Figure 2 As shown, step 103 specifically includes the following steps: Step 301: Divide the substation bay layer into multiple protection zones and configure a corresponding intelligent agent for each protection zone.

[0043] In this step, a protected area refers to an independent control zone defined based on the connection relationships and protection range of electrical equipment within a substation bay. An intelligent agent refers to a logical processing unit configured for each protected area, possessing data processing and strategy generation capabilities, including functional modules that reflect the operational requirements of protection devices within the corresponding protected area.

[0044] In this embodiment of the application, the substation bay layer is divided into multiple independent protection areas according to the power grid topology and equipment protection range. For example, one way to divide it is by feeder lines and busbars. Each divided protection area is configured with a dedicated intelligent agent, so that each intelligent agent is only responsible for generating the parameter adjustment strategy of the protection device in the corresponding protection area, thus obtaining a protection area with intelligent agent configuration.

[0045] Step 302: Define the communication protocol between the intelligent agents, establish an information interaction channel between intelligent agents in adjacent protection areas based on the communication protocol, and transmit the operating status parameters of the substation bay layer protection devices associated with each intelligent agent through the information interaction channel to form a state sharing mechanism between intelligent agents.

[0046] In this step, the communication protocol refers to the standardized protocol established to regulate the data interaction format, interaction frequency, and data verification rules between intelligent agents. It includes syntax, semantics, and timing conventions reflecting information transmission between intelligent agents, and is formulated based on the reliability requirements of power system data transmission and the functional requirements of the intelligent agents. The information interaction channel refers to a dedicated channel established based on the communication protocol for data transmission between intelligent agents in adjacent protection zones. The state sharing mechanism refers to a mechanism that enables real-time sharing of operating status parameters of associated protection devices among each intelligent agent through the information interaction channel. This allows each intelligent agent to obtain the operating status of adjacent areas in real time, avoiding policy conflicts caused by information silos.

[0047] In this embodiment, based on the reliability requirements of power system data transmission, a data interaction format, verification rules, and transmission frequency between intelligent agents are defined to form a communication protocol. Then, based on this communication protocol, a dedicated information interaction channel is built for intelligent agents in adjacent protection areas using the communication network of the substation bay layer. Subsequently, through this information interaction channel, the operating status parameters of the protection device associated with each intelligent agent are transmitted to the adjacent intelligent agents in real time, forming a state sharing mechanism in which each intelligent agent can share the operating status of adjacent areas.

[0048] Step 303: Based on the fault characteristics in the target characteristic waveform data and combined with the bus voltage level in the real-time operating parameters, determine the action constraint range.

[0049] In this step, the action constraint range refers to the set of ranges that limit the adjustment boundaries of the protection device parameters. This range includes the adjustable range of the action threshold and the variable range of the delay parameter, which reflect the safe operation requirements of the protection device.

[0050] In this embodiment, fault features are extracted from the target characteristic waveform data, such as signal duration and amplitude variation range. Combined with the bus voltage level in the real-time operating parameters, the safety boundary for adjusting the protection device parameters is calculated, and the action constraint range, including the adjustable range of the action threshold and the variable range of the delay parameter, is determined.

[0051] Step 304: Based on the real-time running parameters and the action constraint range, calculate the action threshold adjustment direction and delay parameter scaling ratio for each agent through a deep reinforcement learning network to generate a preliminary strategy for each agent.

[0052] In this step, the action threshold adjustment direction refers to the instruction calculated by the agent that the action threshold of the protection device should be adjusted in the direction of increasing or decreasing, including the adjustment tendency reflecting the protection sensitivity optimization requirements under the current operating conditions. The delay parameter scaling ratio refers to the numerical ratio calculated by the agent that the delay parameter of the protection device should be increased or decreased proportionally, including the proportional coefficient reflecting the protection speed and selectivity balance requirements under the current operating conditions. The preliminary strategy refers to the initial adjustment strategy generated by each agent based on its own input data, including the action threshold adjustment direction and the delay parameter scaling ratio, including parameter adjustment content reflecting the independent requirements of the corresponding protection area.

[0053] In this embodiment of the application, step 304 specifically includes the following steps: Step 311: In the policy calculation module configured in each agent, the signal duration of the fault characteristics is converted into time quantization value, the amplitude change range is converted into amplitude quantization value, and the power output value, power flow direction and bus voltage level in the real-time operating parameters are converted into power output quantization value, enumeration value and voltage standard value respectively to form a state quantization set.

[0054] In this step, the strategy calculation module refers to the functional module configured in each intelligent entity for processing input data and generating adjustment strategies, including sub-modules reflecting data quantization, action enumeration, and strategy calculation. Time quantization value refers to the value converted from the signal duration in the fault characteristics to a value recognizable by the deep reinforcement learning network. Amplitude quantization value refers to the value converted from the amplitude variation range in the fault characteristics to a value recognizable by the deep reinforcement learning network. Output quantization value refers to the value converted from the power output value in the real-time operating parameters to a value recognizable by the deep reinforcement learning network. Enumeration value refers to the discrete value converted from the power flow direction in the real-time operating parameters to a value recognizable by the deep reinforcement learning network, obtained based on the actual flow direction of the power flow and preset enumeration rules. Voltage standard value refers to the standardized value converted from the bus voltage level in the real-time operating parameters to a value recognizable by the deep reinforcement learning network, including indicators reflecting the degree of deviation of the bus voltage from the rated value, used as voltage characteristics for network input. State quantization set refers to the dataset formed by integrating time quantization value, amplitude quantization value, output quantization value, enumeration value, and voltage standard value, which can be directly processed by the deep reinforcement learning network.

[0055] In this embodiment, firstly, in the policy calculation module built into each intelligent agent, the signal duration in the fault characteristics is converted into a time quantization value that the network can recognize. For example, 0.5 seconds is converted into a quantization value of 0.5, and the amplitude change range is converted into an amplitude quantization value. For example, 5-15A is converted into an amplitude deviation quantization value of 10. At the same time, the power output value in the real-time operating parameters is converted into an output quantization value. For example, 80MW output is converted into a quantization value of 0.8 relative to the rated 100MW, the power flow direction of the power grid is converted into an enumeration value. For example, the power flow from bus A to bus B is recorded as 1, and the bus voltage level is converted into a voltage standard value. For example, 10.2kV is converted into a standard value of 1.02 relative to the rated 10kV. The above time quantization value, amplitude quantization value, output quantization value, enumeration value, and voltage standard value are integrated to form the state quantization set of each intelligent agent.

[0056] Step 312: Based on the state quantization set, set the adjustment action options that the agent can perform.

[0057] In this step, the adjustment action options refer to the set of protection parameter adjustment types that the agent can execute based on the state quantization set, including forward adjustment of action threshold, reverse adjustment of action threshold, forward scaling of delay parameter, and reverse scaling of delay parameter.

[0058] In this embodiment, based on the current working condition reflected by the state quantization set, four types of adjustment action options that the agent can execute are set, namely, positive adjustment of action threshold, negative adjustment of action threshold, positive scaling of delay parameter, and negative scaling of delay parameter, thereby clarifying the adjustment action range of the agent.

[0059] Step 313: Based on the adjustable range of the action threshold within the action constraint range, set the maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio for the forward adjustment of the action threshold, the reverse adjustment of the action threshold, the forward scaling of the delay parameter, and the reverse scaling of the delay parameter, respectively.

[0060] In this step, the maximum adjustment range refers to the maximum allowable numerical range for both forward and reverse adjustments of the action threshold, including constraint values ​​reflecting the upper limit of the action threshold adjustment. The minimum adjustment range refers to the minimum allowable numerical range for both forward and reverse adjustments of the action threshold, including constraint values ​​reflecting the lower limit of the action threshold adjustment, used to ensure that the action threshold adjustment has an actual optimization effect. The maximum scaling ratio refers to the maximum allowable scaling factor for both forward and reverse scaling of the delay parameter, including constraint values ​​reflecting the upper limit of the delay parameter adjustment. The minimum scaling ratio refers to the minimum allowable scaling factor for both forward and reverse scaling of the delay parameter, including constraint values ​​reflecting the lower limit of the delay parameter adjustment.

[0061] In this embodiment, based on the adjustable range of the action threshold within the action constraint range, a maximum adjustment range and a minimum adjustment range are set for the forward and reverse adjustments of the action threshold, respectively; based on the variable range of the delay parameter within the action constraint range, a maximum scaling ratio and a minimum scaling ratio are set for the forward and reverse scaling of the delay parameter, respectively, thus clarifying the boundary constraints of each adjustment action.

[0062] Step 314: Using deep reinforcement learning, iteratively calculate the state quantization set, maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio to generate the corresponding action value.

[0063] In this step, action value refers to the optimization value score corresponding to each adjustment action option, which is obtained through deep reinforcement learning iterations. This score includes a score reflecting the potential of the action option to improve protection performance under the current operating conditions.

[0064] In this embodiment, the state quantization set, maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio are input into a deep reinforcement learning network. The network performs multiple rounds of iterative calculations for each adjustment action option with the goal of improving the sensitivity and selectivity of the protection device, and outputs the action value corresponding to each option. For example, the action value of positive adjustment of the action threshold is 85, and that of negative adjustment is 60.

[0065] Step 315: Select the action threshold adjustment direction and delay parameter scaling ratio corresponding to the action with the highest action value, and generate a preliminary strategy for the corresponding agent based on the action threshold adjustment direction and delay parameter scaling ratio.

[0066] In this embodiment, the adjustment action option corresponding to the action with the highest value is selected from the action values, and the action threshold adjustment direction and delay parameter scaling ratio in the option are extracted and integrated to form the initial strategy of the corresponding agent.

[0067] Step 305: Based on the state sharing mechanism, transmit each preliminary strategy through the information interaction channel to identify conflicting parameter adjustment contents in each preliminary strategy. Through the interaction and cooperation between agents, correct the conflicting parameter adjustment contents to obtain the agent strategy. In this step, conflicting parameter adjustments refer to contradictory parameter adjustment instructions in the initial strategies of different agents due to a lack of information sharing. Inter-agent collaboration refers to the process by which agents transmit conflicting strategy details and negotiate corrective measures through information exchange channels, including collaborative logic reflecting conflict negotiation rules and correction priorities. Agent strategy refers to the adjustment strategy obtained by each agent after interactive collaboration to correct conflicts, which is consistent with the strategies of neighboring regions, including parameter adjustments reflecting the need for inter-regional coordination.

[0068] In this embodiment of the application, based on the state sharing mechanism, each agent transmits its own generated preliminary strategy to neighboring agents through the information interaction channel; each agent compares the received neighboring preliminary strategies with its own preliminary strategy to identify the conflicting parts of the parameter adjustment content; and then, through the interaction and negotiation between agents, corrects the conflicting parameter adjustment content to obtain the agent strategy of each agent.

[0069] Step 306: Based on the electrical connection relationship of each protection zone, the agent strategies are associated and integrated to form a multi-agent collaborative strategy.

[0070] In this step, electrical connection relationship refers to the physical connection and electrical association between equipment in each protection zone within the substation bay layer, including topology information reflecting power flow and fault propagation paths between zones.

[0071] In this embodiment, the electrical connection relationship between each protection zone of the substation bay layer is obtained, and the integration order of each agent's strategy is determined according to the relationship. The correction strategies of all agents are integrated in series according to the electrical association logic to form a multi-agent collaborative strategy that covers the entire substation bay layer and has no parameter conflicts.

[0072] This application's embodiments achieve precise control of protection parameters by region, avoiding the coarseness of traditional overall setting methods; solve the problem of information silos between intelligent agents through a state sharing mechanism; ensure the safety of parameter adjustment through action constraint range, avoiding exceeding the tolerance range of equipment and power grid; break through the limitations of traditional solutions relying on preset value tables, and can autonomously adapt to complex operating conditions of multi-source grid connection; solve the problem of improper protection coordination that may be caused by independently generated strategies through intelligent agent interaction and cooperation to correct strategy conflicts; and integrate correction strategies according to electrical connection relationships to ensure global consistency of parameter adjustment.

[0073] Step 104: Determine the target candidate action threshold and target candidate delay parameter of the substation bay layer protection device through the multi-agent collaborative strategy.

[0074] In this step, the multi-agent collaborative strategy involves inputting target feature waveform data and multi-source grid-connected real-time operating parameters into a deep reinforcement learning network. The network iteratively learns the adjustment patterns of protection parameters under different operating conditions. Simultaneously, agents are configured for each protection zone in the substation bay layer. Through information exchange between agents, strategy conflicts are corrected, ultimately forming a strategy capable of collaboratively adjusting the parameters of each protection device. The target candidate action threshold and target candidate delay parameter refer to candidate parameters calculated based on the adjustment rules in the multi-agent collaborative strategy, combined with the protection device's action threshold benchmark value and delay parameter benchmark value. These parameters are then determined after filtering by action constraint ranges. The target candidate action threshold is the minimum electrical quantity threshold for triggering the protection device's action, and the target candidate delay parameter is the delay time after the protection device triggers its action. Together, they constitute the core operating parameters of the protection device.

[0075] Step 104 specifically includes the following steps: Step 401: Add different protection area identifiers to the policy content corresponding to the multi-agent collaborative strategy to obtain multiple sets of policy content with added identifiers corresponding to different protection areas.

[0076] In this step, the protection area identifier refers to a number or code used to uniquely distinguish different protection areas within a substation bay layer. It is used to bind the content of the multi-agent collaborative strategy to a specific protection area, avoiding confusion between strategy and area correspondence. The strategy content refers to the specific instruction information included in the multi-agent collaborative strategy, used to guide the adjustment of protection device parameters. The strategy content after adding the identifier refers to the strategy text formed after binding the protection area identifier with the corresponding strategy content.

[0077] In this embodiment, firstly, a protection area identifier is determined to distinguish different protection areas. Then, the strategy content corresponding to each protection area included in the multi-agent collaborative strategy is obtained, such as the action threshold adjustment direction and the delay parameter scaling ratio. Each protection area identifier is bound to the corresponding strategy content, that is, the identifier of the protection area to which it belongs is added before or at the beginning of each strategy content, so that each strategy content can be clearly associated with a specific protection area. Finally, multiple sets of strategy content with added identifiers are obtained, each matching a different protection area.

[0078] Step 402: Extract the action threshold adjustment command and delay parameter modification command of the substation bay layer protection device in the corresponding protection area from each added identifier strategy content.

[0079] In this step, the action threshold adjustment instruction refers to the instruction extracted from the strategy content after adding the identifier, which guides the adjustment of the action threshold of the protection device. It includes information such as the adjustment direction and adjustment range of the action threshold, reflecting the optimization requirements of different protection zones for the action threshold. The delay parameter modification instruction refers to the instruction extracted from the strategy content after adding the identifier, which guides the adjustment of the delay parameter of the protection device. It includes information such as the modification direction and scaling range of the delay parameter.

[0080] In this embodiment, the structure of each set of strategy content after adding the identifier is first defined, which includes two parts: protection area identifier and strategy text. The strategy text records the parameter adjustment requirements of the corresponding protection area in the multi-agent collaborative strategy in the form of natural language or structured instructions. For a single set of strategy content after adding the identifier, the protection area corresponding to the strategy text is first locked according to the protection area identifier at the top, and then the strategy text is parsed line by line to locate the statements related to the action threshold of the substation bay layer protection device.

[0081] Then, in the located action threshold-related statements, further key information is extracted: expressions such as "positive adjustment," "reverse adjustment," "increase," and "decrease" are identified to determine the adjustment direction; numerical or range expressions such as 0.2A, 0.5 times, and not exceeding 1A are extracted to determine the adjustment magnitude. The above are integrated into a complete action threshold adjustment instruction, for example, a positive action threshold adjustment of 0.3A and a reverse action threshold adjustment not exceeding 0.2A. Subsequently, statements related to the delay parameters of substation bay layer protection devices are located in the same strategy text. Key information is also extracted, including expressions such as "identify," "positive scaling," "reverse scaling," "extend," and "shorten" to determine the modification direction; numerical or range expressions such as 0.05 seconds, 1.2 times, and not less than 0.1 seconds are extracted to determine the modification ratio / duration. The above are integrated into a complete delay parameter modification instruction. This process is repeated to complete the parsing of all strategy content after adding identifiers, ensuring that each protection zone corresponds to a unique action threshold adjustment instruction and delay parameter modification instruction.

[0082] Step 403: Based on the protection area identification, summarize the action threshold reference value and delay parameter reference value of the substation bay layer protection device in each protection area to establish a reference parameter table with area identification.

[0083] In this step, the action threshold reference value refers to the initial action threshold value of the protection device under factory settings or historical optimal operating conditions, obtained based on the protection device's factory parameter manual or historical optimal parameter records. The delay parameter reference value refers to the initial delay parameter value of the protection device under factory settings or historical optimal operating conditions, obtained based on the protection device's factory parameter manual or historical optimal parameter records. The reference parameter table with area identification refers to a table compiled and summarized in the format of protection area identification - protection device number - action threshold reference value - delay parameter reference value.

[0084] In this embodiment, firstly, all defined protection zone identifiers are identified, and a unique device number is assigned to the substation bay layer protection device within each protection zone to ensure accurate location of each device using the protection zone identifier and device number. Next, basic parameters for each protection device are collected. Then, using the protection zone identifier as the core classification dimension, the parameters are entered into a preset table template according to a hierarchical structure of [protection zone identifier, device number, action threshold baseline value, delay parameter baseline value]. The first column of the table is set to the protection zone identifier, the second column to the device number, the third column to the action threshold baseline value, and the fourth column to the delay parameter baseline value. Finally, the parameters in the table are cross-validated, and after confirmation, a baseline parameter table with zone identifiers is formed.

[0085] Step 404: Calculate the action threshold adjustment amount and delay parameter modification amount for each substation bay layer protection device according to the action threshold adjustment instruction, the delay parameter modification instruction, and the reference parameter table.

[0086] In this step, the action threshold adjustment amount refers to the specific value that the protection device's action threshold needs to be adjusted, calculated according to the action threshold adjustment rules and the action threshold reference value. It reflects the change in the action threshold from the reference value to the optimized value and is associated with the fault characteristics of the corresponding protection area. The delay parameter modification amount refers to the specific value that the protection device's delay parameter needs to be modified, calculated according to the delay parameter modification rules and the delay parameter reference value. The power flow direction refers to the specific direction in which power flows from the power source side to the load side or between different power sources in the power grid.

[0087] In this embodiment, firstly, for each protection zone identifier, the fault characteristics of that zone previously extracted from the target feature waveform data are retrieved, and simultaneously, the power flow direction corresponding to that zone in the power grid is retrieved. For example, the power flow direction of zone 1 is [distributed power source A, main grid bus B]. For the action threshold adjustment amount: firstly, the protection zone identifier and action threshold reference value corresponding to the current protection device are found from the reference parameter table with zone identifiers; then, the action threshold adjustment command corresponding to the protection zone identifier is matched. For example, the action threshold adjustment command for zone 1 is a positive adjustment with an adjustment range of 0.2-0.5A.

[0088] Then, the adjustment range is determined based on the fault characteristics of the area. If the amplitude variation range in the fault characteristics is large, such as the amplitude variation range of 4-16A in area 1, spanning 12A, it indicates that the electrical quantity fluctuates violently during the fault, and the adjustment range needs to be appropriately increased to ensure protection sensitivity. Therefore, a larger value is taken within the command range. If the amplitude variation range is small, such as the amplitude variation range of 6-10A in a certain area, spanning 4A, then a smaller value is taken within the command range. Finally, the calculation is: Action threshold adjustment amount = value corresponding to the command adjustment direction × adjustment range.

[0089] For the delay parameter modification amount: First, find the protection zone identifier and delay parameter benchmark value corresponding to the current protection device from the benchmark parameter table; then match the delay parameter modification instruction corresponding to the protection zone identifier. For example, the delay parameter modification instruction for zone 1 is positive scaling, with a ratio of 1.1-1.3 times; then adjust the ratio according to the power flow direction of the zone. If the power flow direction is [distributed power source, main grid], the output fluctuation of the distributed power source may cause the power flow to reverse, so the scaling ratio needs to be appropriately reduced to avoid excessive delay causing failure to operate. Therefore, take the smaller value within the instruction ratio range; if the power flow direction is [main grid, load], the power flow direction is stable, and the larger value within the instruction ratio range can be taken; finally, calculate: delay parameter modification amount = delay parameter benchmark value × (scaling ratio - 1). Repeat the above process to complete the calculation of the action threshold adjustment amount and delay parameter modification amount for all protection devices.

[0090] Step 405: Superimpose the action threshold adjustment amount with the corresponding action threshold baseline value to obtain the candidate action threshold; superimpose the delay parameter modification amount with the corresponding delay parameter baseline value to obtain the candidate delay parameter; and select the target candidate action threshold and target candidate delay parameter that meet the action constraint range.

[0091] In this step, the candidate action threshold refers to the candidate optimized value of the protection device's action threshold obtained by superimposing the action threshold adjustment amount with the corresponding action threshold baseline value. The candidate delay parameter refers to the candidate optimized value of the protection device's delay parameter obtained by superimposing the delay parameter modification amount with the corresponding delay parameter baseline value. The action constraint range refers to the set of ranges that limit the adjustment boundaries of the protection device's parameters, which includes the adjustable range of the action threshold and the variable range of the delay parameter.

[0092] In this embodiment of the application, firstly, for each substation bay layer protection device, its action threshold reference value is extracted from the reference parameter table with area identification, and the value is superimposed with the device's action threshold adjustment amount. If the adjustment amount is positive, the reference value is added to the adjustment amount; if the adjustment amount is negative, the reference value is subtracted from the adjustment amount to obtain the candidate action threshold of the device.

[0093] Then, the baseline value of the delay parameter of the device is extracted and numerically superimposed with the modification amount of the delay parameter. If the modification amount is positive, the baseline value is added to the modification amount; if the modification amount is negative, the baseline value is subtracted from the modification amount to obtain the candidate delay parameter of the device. Next, the target candidate parameter is screened: the action constraint range is retrieved, and the candidate action threshold of each device is compared with the adjustable range of the action threshold. If the candidate action threshold falls within the adjustable range, the candidate action threshold meets the requirements; if it exceeds the adjustable range, the process returns to step 404 to readjust the action threshold adjustment amount. Similarly, the candidate delay parameter is compared with the variable range of the delay parameter. If the candidate delay parameter falls within the adjustable range, it meets the requirements; if it exceeds the adjustable range, the process returns to step 404 to readjust the delay parameter modification amount. Finally, all candidate action thresholds and candidate delay parameters that meet the action constraint range are determined as the target candidate action threshold and target candidate delay parameter of the device, respectively.

[0094] This application's embodiments achieve precise binding between strategy content and protection areas, avoiding the problem of confusing parameter and area correspondence in traditional schemes; enable parameter adjustment to adapt to the fault characteristics and power flow changes of different areas; and prevent parameter adjustment from exceeding the safety boundary by comparing with the action constraint range to screen target candidate parameters, thus solving the problem of improper coordination caused by the lack of parameter safety verification in existing schemes.

[0095] Step 105: Using digital twins, construct a virtual substation system with the same structure as the actual substation. Combine the target candidate action threshold and target candidate delay parameters to simulate the action response records of the substation bay layer protection device under various fault scenarios. Based on the comparison results of the action response records with the preset response standard set, generate the action verification conclusion of the adaptive setting process.

[0096] In this step, the virtual substation system refers to a virtual system constructed in a virtual environment using digital twin technology, based on the physical structural dimensions, protection device installation locations, and electrical connection parameters of an actual substation, perfectly matching the structure and electrical connections of the actual substation. The fault scenario refers to a simulated fault situation generated based on the target fault type, including the fault trigger location, fault duration, and fault characteristic waveforms. The action response record refers to the action information recorded for each virtual protection device when simulating fault scenarios in the virtual substation system. This record includes the action initiation time, action execution type, and actual parameter values, reflecting the actual action behavior of the protection device under the fault scenario. The preset response standard set refers to a set of standard action requirements formulated based on the four characteristics of relay protection for different fault scenario numbers. This set includes the standard action time interval, standard action type, and standard parameter value range, used as the basis for judging whether the protection device's action is qualified. The comparison result refers to the result obtained by comparing the action response records under the same scenario number with the preset response standard set item by item. The adaptive tuning process refers to the entire process in this technical solution, from collecting historical fault data, extracting features, generating protection strategies, to determining the target candidate action threshold and delay parameters. It can autonomously adjust the protection device parameters according to the real-time operating status of the power grid, without relying on a fixed setpoint table, and adapt to dynamically changing system conditions. The action verification conclusion refers to the conclusion drawn from comparing the action response records under all fault scenarios with the preset response standard set, comprehensively judging whether the target candidate action threshold and delay parameters meet the protection requirements.

[0097] In this embodiment of the application, step 105 specifically includes the following steps: Step 501: Based on the physical structure dimensions of the actual substation, the installation location of the substation bay layer protection device, and the electrical connection line parameters, construct a virtual substation system, assign different virtual identifiers to each virtual protection device in the virtual substation system, and establish a correspondence table between the virtual identifiers and the substation bay layer protection devices.

[0098] In this step, electrical connection line parameters refer to a set of parameters reflecting the electrical characteristics and connection relationships of lines within an actual substation. These parameters include the physical length of the line, conductor cross-sectional area, line impedance, insulation class, and information on the connection equipment at both ends of the line. A virtual substation system refers to a virtual system constructed in a virtual environment that is completely identical in structure to an actual substation. A virtual protection device refers to a virtual unit in the virtual substation system that simulates the function and parameters of the bay-level protection devices in an actual substation. A virtual identifier refers to a unique number assigned to each virtual protection device, based on the device's location and numbering rules. A correspondence table records the one-to-one correspondence between virtual identifiers and actual substation bay-level protection devices, including the association information of virtual identifier - actual protection device number - associated protection area.

[0099] In this embodiment, the physical structural dimensions of the actual substation, the installation location of the substation bay layer protection device, and the electrical connection line parameters are first collected. Then, based on the physical structural dimensions, the relative positions of the equipment and the channel width are set in the virtual environment to form a virtual space layout. According to the installation location of the protection device, virtual protection devices are deployed at corresponding points in the virtual space layout. According to the electrical connection line parameters, virtual connection lines with the same impedance and connection relationship are built between the virtual protection device and the virtual busbar to integrate and form a virtual substation system. Subsequently, each virtual protection device is assigned a unique virtual identifier, and a correspondence table of virtual identifier - actual protection device number - protection area is established.

[0100] Step 502: According to the correspondence table, match the target candidate action threshold and the target candidate delay parameter to the virtual identifier in the corresponding protection area to form a set of virtual protection devices.

[0101] In this step, the virtual protection device set refers to the group of virtual protection devices with complete adjustment parameters formed after matching the target candidate action threshold and the target candidate delay parameter to the corresponding virtual identifier.

[0102] In this embodiment, the corresponding relationship table and the target candidate action threshold and target candidate delay parameter are retrieved. Based on the association between the virtual identifier and the protection area to which it belongs in the corresponding relationship table, the target candidate action threshold and target candidate delay parameter of the same protection area are matched to the virtual protection device with the corresponding virtual identifier, so that each virtual protection device is loaded with the optimized parameters, forming a set of virtual protection devices.

[0103] Step 503: Generate multiple sets of simulated fault scenario parameters according to the target fault type, and assign different scenario numbers to each set of simulated fault scenario parameters.

[0104] In this step, the simulated fault scenario parameters refer to the set of parameters generated to test the reliability of the protection device parameters, simulating the characteristics of actual faults. These parameters include the fault trigger location, fault duration, and fault characteristic waveforms. The scenario number refers to a unique identifier assigned to each set of simulated fault scenario parameters, which is obtained based on the generation order of the simulated fault scenario parameters.

[0105] In this embodiment of the application, multiple sets of simulated fault scenario parameters are generated according to the target fault type, such as three-phase short circuit or single-phase ground fault. Each set of parameters includes the fault trigger location, fault duration, and fault characteristic waveform. Different scenario numbers are assigned to each set of simulated fault scenario parameters in sequence to ensure that each set of parameters corresponds to a unique number.

[0106] Step 504: In the virtual substation system, simulate a virtual line fault according to the scenario number, and record the action response record of each virtual protection device after the virtual line fault is triggered.

[0107] In this step, virtual line fault refers to a virtual event in the virtual substation system where a current and voltage signal corresponding to a fault characteristic waveform is applied to the target virtual line to simulate an actual line fault.

[0108] In this embodiment of the application, step 504 may specifically include the following steps: Step 511: Based on the scenario number, locate the target virtual line in the line layout diagram of the virtual substation system that corresponds to the fault triggering location in the simulated fault scenario parameters.

[0109] In this step, the circuit layout diagram refers to the topology diagram in the virtual substation system that shows the location relationships between all virtual connection lines and virtual protection devices. The target virtual line refers to the virtual line in the virtual substation system that corresponds to the fault trigger location in the simulated fault scenario parameters.

[0110] In this embodiment of the application, based on the scenario number, a virtual line corresponding to the fault triggering position in the simulation fault scenario parameters of the scenario is found in the line layout diagram, and it is determined as the target virtual line.

[0111] Step 512: Based on the fault duration in the simulated fault scenario parameters, set the simulation running time, combine the current signal and voltage signal corresponding to the fault characteristic waveform, trigger the virtual circuit fault of the target virtual circuit, and record the signal change value of the target virtual circuit.

[0112] In this step, the simulation runtime refers to the time set from fault triggering to fault termination in the virtual line fault simulation. The signal change value refers to the real-time values ​​of the current and voltage signals on the target virtual line and related lines as a function of time after the virtual line fault is triggered.

[0113] In this embodiment, the simulation running time is set based on the fault duration in the simulated fault scenario parameters. Current and voltage signals corresponding to the fault characteristic waveforms are applied to the target virtual line to trigger the virtual line fault, and the signal change values ​​of the target virtual line and associated lines are recorded in real time.

[0114] Step 513: Compare the signal change value with the target candidate action threshold, and record the time point when the signal change value exceeds the target candidate action threshold as the action start time, the preset protection operation performed by the virtual protection device as the action execution type, and the candidate action threshold and candidate delay parameter used by the virtual protection device when performing the preset protection operation as the actual parameter values.

[0115] In this step, the action initiation time refers to the specific point in time when the virtual protection device detects a signal change value exceeding the target candidate action threshold. The preset protection operation refers to the protection actions pre-set within the virtual protection device, corresponding to different fault types. The action execution type refers to the type of preset protection operation actually executed by the virtual protection device after a fault is triggered. The actual parameter values ​​refer to the candidate action threshold and candidate delay parameters actually used by the virtual protection device when executing the preset protection operation.

[0116] In this embodiment, the recorded signal change value is compared with the target candidate action threshold of the virtual protection device after loading and adjusting parameters in real time. When the signal change value exceeds the threshold, the time point at this time is recorded as the action start time, the preset protection operation performed by the virtual protection device is recorded as the action execution type, and the target candidate action threshold and delay parameter used when performing the operation are recorded as the actual parameter values.

[0117] Step 514: After the virtual line fault that triggered the target virtual line ends, the action start time, action execution type and actual parameter value of each virtual protection device are integrated according to the virtual identifier to form an action response record with associated scenario number.

[0118] In this embodiment of the application, after the simulation running time ends and the virtual line fault stops, the action start time, action execution type, and actual parameter values ​​of each virtual protection device are integrated according to the virtual identifier in the format of virtual identifier-scenario number-response information to form an action response record associated with the scenario number.

[0119] Step 505: Based on the four characteristics of relay protection and the scenario number, determine the preset response standard set.

[0120] In this step, the four criteria for relay protection refer to the four core standards for evaluating the performance of relay protection devices. These criteria include selectivity, speed, sensitivity, and reliability. The preset response standard set refers to a set of standards developed based on the four criteria for relay protection, numbered for different scenarios, to evaluate whether the virtual protection device's operation is qualified. This standard set includes the standard operation time interval, standard operation type, and standard parameter value range.

[0121] In this embodiment, the four criteria of relay protection are retrieved, and the simulated fault scenario parameters corresponding to each scenario number are combined to formulate a preset response standard set for each scenario number: a standard action time interval is set for the speed requirement, such as 0.05-0.1 seconds after the fault is triggered; a standard action type is set for the selectivity requirement, such as tripping for short circuit faults; and a standard parameter value range is set for the sensitivity requirement, so as to ensure that each scenario number corresponds to a unique preset response standard set.

[0122] Step 506: Compare the action response records under the same scene number with the preset response standard set to form the comparison results corresponding to each scene number, so as to generate the action verification conclusion of the adaptive tuning process based on the comparison results.

[0123] In this embodiment, the action response records under the same scene number are compared item by item with the preset response standard set: it is determined whether the action start time is within the standard range, whether the action execution type is consistent with the standard, and whether the actual parameter values ​​meet the standard range, thus forming the comparison result of the scene number; combining the comparison results of all scene numbers, if all scenes meet the standard, a parameter tuning qualified action verification conclusion is generated; if there are scenes that do not meet the standard, a conclusion that parameters in a certain area need to be optimized is generated, thus completing the action verification of the adaptive tuning process.

[0124] The embodiments of this application can autonomously adapt to complex and ever-changing multi-source grid-connected operating conditions, solving the problem of limited adaptability of protection devices under extreme or atypical operating conditions; reducing computational overhead and time consumption, realizing real-time closed-loop protection setting and verification, timely detecting potential mismatches or action delays caused by parameter adjustments, and ultimately ensuring that the protection device can maintain high sensitivity and selectivity under different operating conditions, avoiding false tripping or failure to trip, and improving the adaptive capability and overall reliability of the intelligent substation protection system.

[0125] This application collects historical fault current / voltage waveform data and real-time grid operating parameters of the power system using fault recorders, current transformers, and voltage transformers. It also collects operating status parameters of protection devices using edge computing nodes at the substation bay level, and simultaneously obtains the actual physical structure of the substation, the installation location of protection devices, and electrical connection line parameters. The collected data is used to extract fault features, generate multi-agent collaborative strategies to determine target candidate parameters for protection devices, and finally, a virtual substation system constructed using digital twins simulates fault scenarios to verify the effectiveness of parameter settings.

[0126] Figure 3 This is a schematic diagram of a specific implementation of the adaptive tuning and action verification system for a spacer protection device provided in this application, referring to... Figure 3 The system may include: The acquisition module 21 is used to acquire historical fault current and voltage waveform data of the power system, and extract the current characteristic waveform and voltage characteristic waveform corresponding to the target fault type from the historical fault current and voltage waveform data. Processing module 22 is used to preprocess and compress the current characteristic waveform and voltage characteristic waveform through edge computing nodes configured in the substation bay layer to obtain target characteristic waveform data; The generation module 23 is used to generate a multi-agent cooperative strategy for the substation bay layer protection device through a deep reinforcement learning network based on the real-time operating parameters of the power grid and the target feature waveform data. The real-time operating parameters include power output value, power grid flow direction and bus voltage level. The determination module 24 is used to determine the target candidate action threshold and target candidate delay parameters of the substation bay layer protection device through the multi-agent collaborative strategy; The construction module 25 is used to construct a virtual substation system with the same structure as the actual substation using digital twins. It combines the target candidate action threshold and the target candidate delay parameter to simulate the action response records of the substation bay layer protection device under various fault scenarios. Based on the comparison results of the action response records with the preset response standard set, it generates the action verification conclusion of the adaptive setting process.

[0127] The adaptive setting and operation verification system for the spacer protection device in this application is used to implement the aforementioned adaptive setting and operation verification method for the spacer protection device. Therefore, the specific implementation of the adaptive setting and operation verification system for the spacer protection device can be found in the embodiment section of the adaptive setting and operation verification method for the spacer protection device above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0128] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the adaptive tuning and action verification method for any of the above-described spacer protection devices.

[0129] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the adaptive tuning and action verification method for any of the above-described methods for spacer protection devices.

[0130] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0131] The embodiments of this application also provide a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps in the embodiments of the adaptive tuning and action verification method for any type of spacer protection device.

[0132] Those skilled in the art will further 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 implementation should not be considered beyond the scope of this application.

[0133] The adaptive setting and operation verification method and system for spacer protection devices provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An adaptive setting and operation verification method for bay layer protection devices, characterized in that, include: Collect historical fault current and voltage waveform data of the power system, and extract the current characteristic waveform and voltage characteristic waveform corresponding to the target fault type from the historical fault current and voltage waveform data; By configuring edge computing nodes in the substation bay layer, the current characteristic waveform and voltage characteristic waveform are preprocessed and compressed to obtain target characteristic waveform data. Based on the real-time operating parameters of the power grid and the target characteristic waveform data, a multi-agent collaborative strategy for the substation bay layer protection device is generated through a deep reinforcement learning network. The real-time operating parameters include power output value, power grid flow direction and bus voltage level. The target candidate action threshold and target candidate delay parameter of the substation bay layer protection device are determined by the multi-agent cooperative strategy. Using digital twins, a virtual substation system with the same structure as the actual substation is constructed. Combining the target candidate action threshold and the target candidate delay parameter, the action response records of the substation bay layer protection device are simulated under various fault scenarios. Based on the comparison results of the action response records with the preset response standard set, the action verification conclusion of the adaptive setting process is generated.

2. The method according to claim 1, characterized in that, Based on the real-time operating parameters of the power grid and the target characteristic waveform data, a multi-agent cooperative strategy for the substation bay layer protection device is generated through a deep reinforcement learning network, including: The substation bay layer is divided into multiple protection zones, and a corresponding intelligent agent is configured for each protection zone. Define a communication protocol between the agents, establish an information interaction channel between agents in adjacent protection zones based on the communication protocol, and transmit the operating status parameters of the substation bay layer protection devices associated with each agent through the information interaction channel to form a state sharing mechanism between agents. Based on the fault characteristics in the target characteristic waveform data, and combined with the bus voltage level in the real-time operating parameters, the action constraint range is determined. Based on the real-time operating parameters and the action constraint range, the action threshold adjustment direction and delay parameter scaling ratio corresponding to each agent are calculated through a deep reinforcement learning network to generate the preliminary strategy corresponding to each agent. Based on the state sharing mechanism, each preliminary strategy is transmitted through the information interaction channel to identify conflicting parameter adjustment contents in each preliminary strategy. Through the interaction and cooperation between agents, the conflicting parameter adjustment contents are corrected to obtain the agent strategy. Based on the electrical connection relationship of each protected area, the agent strategies are correlated and integrated to form a multi-agent collaborative strategy.

3. The method according to claim 2, characterized in that, Based on the real-time operating parameters and the action constraint range, a deep reinforcement learning network is used to calculate the action threshold adjustment direction and delay parameter scaling ratio for each agent, in order to generate a preliminary strategy for each agent, including: In the policy calculation module configured in each agent, the signal duration of fault characteristics is converted into time quantization value, the amplitude change range is converted into amplitude quantization value, and the power output value, power flow direction and bus voltage level in real-time operating parameters are converted into power output quantization value, enumeration value and voltage standard value respectively to form a state quantization set. Based on the state quantization set, the adjustment action options that the agent can perform are set; Based on the adjustable range of the action threshold within the action constraint range, the maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio are set for the forward adjustment of the action threshold, the reverse adjustment of the action threshold, the forward scaling of the delay parameter, and the reverse scaling of the delay parameter, respectively. Using deep reinforcement learning, the state quantization set, maximum adjustment range, minimum adjustment range, maximum scaling ratio, and minimum scaling ratio are iteratively calculated to generate the corresponding action value. Select the action threshold adjustment direction and delay parameter scaling ratio corresponding to the action with the highest action value, and generate a preliminary strategy for the corresponding agent based on the action threshold adjustment direction and delay parameter scaling ratio.

4. The method according to claim 1, characterized in that, The target candidate action threshold and target candidate delay parameters of the substation bay layer protection device are determined through the multi-agent cooperative strategy, including: Different protection zone identifiers are added to the policy content corresponding to the multi-agent collaborative strategy to obtain multiple sets of policy content with added identifiers corresponding to different protection zones. Extract the action threshold adjustment command and delay parameter modification command of the substation bay layer protection device in the corresponding protection area from each added flag strategy content; Based on the protection area identification, the reference values ​​of the action threshold and delay parameters of the substation bay layer protection devices in each protection area are summarized to establish a reference parameter table with area identification. Based on the action threshold adjustment instruction, the delay parameter modification instruction, and the reference parameter table, calculate the action threshold adjustment amount and delay parameter modification amount of each substation bay layer protection device; The action threshold adjustment amount is superimposed with the corresponding action threshold baseline value to obtain the candidate action threshold. The delay parameter modification amount is superimposed with the corresponding delay parameter baseline value to obtain the candidate delay parameter. The target candidate action threshold and target candidate delay parameter that meet the action constraint range are selected.

5. The method according to claim 1, characterized in that, Using digital twins, a virtual substation system with the same structure as the actual substation is constructed. Combining the target candidate action threshold and target candidate delay parameters, the action response records of the substation bay layer protection devices are simulated under various fault scenarios. Based on the comparison results between the action response records and a preset response standard set, action verification conclusions for the adaptive setting process are generated, including: Based on the physical structure dimensions of the actual substation, the installation location of the substation bay layer protection device, and the electrical connection line parameters, a virtual substation system is constructed, and each virtual protection device in the virtual substation system is assigned a different virtual identifier. A correspondence table between the virtual identifier and the substation bay layer protection device is established. According to the correspondence table, the target candidate action threshold and the target candidate delay parameter are matched to the virtual identifiers in the corresponding protection area to form a set of virtual protection devices; Based on the target fault type, generate multiple sets of simulated fault scenario parameters and assign different scenario numbers to each set of simulated fault scenario parameters; In the virtual substation system, a virtual line fault is simulated according to the scenario number, and the action response record of each virtual protection device after the virtual line fault is triggered is recorded. Based on the four characteristics of relay protection and the scenario number, a set of preset response standards is determined; The action response records under the same scene number are compared with the preset response standard set to form the comparison results corresponding to each scene number, so as to generate the action verification conclusion of the adaptive tuning process based on the comparison results.

6. The method according to claim 5, characterized in that, In the virtual substation system, a virtual line fault is simulated according to the scenario number, and the action response record of each virtual protection device after the virtual line fault is triggered is recorded, including: Based on the scenario number, locate the target virtual line in the line layout diagram of the virtual substation system that corresponds to the fault triggering position in the simulated fault scenario parameters; Based on the fault duration in the simulated fault scenario parameters, the simulation runtime is set, and combined with the current and voltage signals corresponding to the fault characteristic waveform, the virtual circuit fault of the target virtual circuit is triggered, and the signal change value of the target virtual circuit is recorded. The signal change value is compared with the target candidate action threshold. The time point when the signal change value exceeds the target candidate action threshold is recorded as the action start time, the preset protection operation performed by the virtual protection device is recorded as the action execution type, and the candidate action threshold and candidate delay parameter used by the virtual protection device when performing the preset protection operation are recorded as the actual parameter values. After the virtual line fault that triggered the target virtual line is terminated, the action start time, action execution type and actual parameter values ​​of each virtual protection device are integrated according to the virtual identifier to form an action response record with associated scenario number.

7. The method according to claim 1, characterized in that, By configuring edge computing nodes in the substation bay layer, the current characteristic waveform and voltage characteristic waveform are preprocessed and compressed to obtain target characteristic waveform data, including: Using the time of the fault occurrence as the time reference point, the preset waveform segments in the current characteristic waveform and voltage characteristic waveform are extracted by the edge computing nodes configured in the substation bay layer. The preset waveform segments are divided into multiple initial waveform segments, and a timestamp is added to each initial waveform segment to obtain the intermediate waveform segments. Compare the rate of change of signal amplitude between adjacent intermediate waveform segments to identify abnormal waveform segments whose rate of change of signal amplitude exceeds a preset range; Using the peak and valley points of the signal amplitude in the abnormal waveform segment as boundaries, a target waveform segment is extracted from the abnormal waveform segment; The target waveform segment is subjected to signal denoising processing to remove high-frequency interference signals, resulting in a denoised waveform segment. The denoised waveform segment is decomposed into sub-waveform components in different frequency ranges, where each sub-waveform component corresponds to a frequency label. Based on the frequency label, the signal duration and amplitude variation range of the key sub-waveform components are statistically analyzed; Based on the timestamp identifier, the key sub-waveform components, the signal duration, and the amplitude variation range are integrated to generate target feature waveform data.

8. An adaptive setting and action verification system for bay layer protection devices, characterized in that, include: The acquisition module is used to acquire historical fault current and voltage waveform data of the power system, and extract the current characteristic waveform and voltage characteristic waveform corresponding to the target fault type from the historical fault current and voltage waveform data; The processing module is used to preprocess and compress the current characteristic waveform and voltage characteristic waveform through edge computing nodes configured in the substation bay layer to obtain target characteristic waveform data. The generation module is used to generate a multi-agent cooperative strategy for the substation bay layer protection device through a deep reinforcement learning network based on the real-time operating parameters of the power grid and the target feature waveform data. The real-time operating parameters include power output value, power grid flow direction and bus voltage level. The determination module is used to determine the target candidate action threshold and target candidate delay parameters of the substation bay layer protection device through the multi-agent collaborative strategy; The module is used to construct a virtual substation system with the same structure as the actual substation using digital twins. It combines the target candidate action threshold and the target candidate delay parameter to simulate the action response records of the substation bay layer protection device under various fault scenarios. Based on the comparison results of the action response records with the preset response standard set, it generates the action verification conclusion of the adaptive setting process.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the adaptive tuning and operation verification method for a spacer protection device as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the adaptive tuning and action verification method for a spacer protection device as described in any one of claims 1 to 7.

Images