A verification method for an intelligent integrated protection device

Abstract

This invention relates to the field of protection device testing technology, specifically to a verification method for an intelligent integrated protection device. In this invention, the start time and response time in the tripping state transition sequence of the protection device are extracted, execution delay and action segment length are calculated, and abnormal links are identified based on the state number sequence comparison. This achieves the parsing and standardized reconstruction of the state logic structure, forming an ordered tripping path. A long short-term memory neural network is introduced to align the state change sequence with the current signal segment in time. Through the linkage analysis of state-current abrupt change positions, potential mismatches or abnormal change trends in the state chain are identified, and the state path is reconstructed, enhancing path continuity and dynamic adaptability. A random forest is used to train and model the device, using the device number, number of jump segments, and response residual as multiple inputs. An ensemble tree voting mechanism is used to output device scores, achieving the identification and ranking of devices with significant abnormal frequencies.

Description

Technical Field

[0001] This invention relates to the field of protection device testing technology, and in particular to a verification method for an intelligent integrated protection device. Background Technology

[0002] The key to the field of protection device testing technology lies in systematically and standardizedly verifying and evaluating the functions and performance of various protection devices in power systems. This ensures that the devices can act promptly and accurately according to predetermined logic under abnormal conditions such as faults, voltage fluctuations, and short circuits, thereby effectively isolating faults, protecting the safe operation of equipment, and maintaining the stability of the power grid.

[0003] A verification method for an intelligent integrated protection device aims to comprehensively test the response behavior of the protection device under simulated faults, normal operation, and boundary conditions. The purpose is to determine whether the action of the protection device is consistent with the set parameters, verify the correctness and timeliness of the response, identify potential malfunctions or failures to operate, improve the reliability and fault tolerance of the protection device in the actual power grid environment, enhance the system's security defense capabilities, and ensure the timeliness and accuracy of fault handling.

[0004] Existing technologies, in the process of verifying protection devices, lack in-depth modeling of the state logic chain and sequence response flow. Fault response behavior is judged based on start and end states or delay parameters, which is insufficient to cover structural anomalies such as jumps, divergences, and sequence disorder that may occur during state transitions. Under the multi-state driven and multi-condition blocking logic of protection devices, traditional verification methods struggle to distinguish between seemingly normal and internally abnormal behavior patterns. The abrupt changes in current response are not fully quantified and analyzed, and the coupling changes between state and electrical quantity lack a linkage identification mechanism, easily overlooking short-term, non-continuous malfunctions. The judgment of response results is mostly based on single events, lacking systematic collection and statistical ranking of similar devices and similar anomalies, making it difficult to identify and warn of potentially high-risk devices. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a verification method for an intelligent integrated protection device.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a verification method for an intelligent integrated protection device, comprising the following steps:

[0007] S1: Based on the event change sequence and switch status record recorded during the operation of the protection device, the start time and trip response time are extracted, the execution delay and action duration are calculated, and the status number is compared for sequential consistency to obtain the trip logic transfer path set.

[0008] S2: Based on the trip logic transfer path set, filter the transfer nodes in the interlocking logic module that have the same trigger condition value but different state successors, extract the differences in jump time and path structure, and generate a state divergence point map set.

[0009] S3: Based on the state divergence point map set, a long short-term memory neural network is used to extract the state change sequence and current data fragment in the alarm triggering logic, perform state sequence comparison and mutation detection, insert reconstruction fragments to form a sequence path flow, and obtain abnormal state reconstruction path group;

[0010] S4: Based on the abnormal state, reconstruct the path group, identify the position of the jump segment in the regression response mechanism, calculate the response residual and judge the over-limit segment, use random forest to locate the state starting point corresponding to the jump segment and classify and summarize it to obtain the jump behavior risk index cluster.

[0011] S5: Based on the jump behavior risk index cluster, extract the device number and time information corresponding to the jump segment, count the response residual and the number of jumps, score and sort, and establish a risk sequence table of high-frequency jump devices.

[0012] As a further aspect of the present invention, the specific steps for generating the trip logic transfer path set are as follows:

[0013] Based on the event change sequence and switch status record recorded during the operation of the protection device, the time of the first and last nodes in each sequence is read. By performing the difference calculation between the start time and the completion time and uniformly numbering the node labels, the time unit of each record is checked and converted into a unified reference time format. After removing abnormal time records and completing the cleaning, the duration within the tripping section is calculated, and a tripping response time parameter group is generated.

[0014] Based on the trip response time parameter group, the status numbers of each segment are sorted. The index difference of the consecutive number sequence is compared item by item to determine whether it is the minimum single step value. Reverse matching verification is performed. Numbers with gaps are marked and offset compensation is performed. The number stream that passes the consistency check is structured and organized to generate a status number sequence structure set.

[0015] Based on the state number sequence structure set, the number sequence is bound to the time parameter, each group of numbers is sequentially corresponding to the response time period and rearranged into a time stream, the action triggering stage and the recovery stage in the time stream are divided, the sequence is spliced ​​and combined in segments and the action chain is reorganized according to the triggering order to obtain the trip logic transfer path set.

[0016] As a further aspect of the present invention, the specific steps for generating the state divergence point map set are as follows:

[0017] Based on the trip logic transfer path set, the input trigger conditions of all state segments in the interlocking logic module are extracted, grouped according to the input condition values, and state pairs with identical input values ​​are selected. All subsequent state information of the state pairs are obtained and type-labeled. State pairs with structural differences are identified and grouped into the same set to generate a trigger condition consistent branch group.

[0018] Based on the consistent branching group of the triggering conditions, the difference between the start time difference and the response time span of each state segment is calculated. The correspondence between the triggering time and the sequential number of the jump node is matched. The path length between each pair of nodes is calculated and their jump direction and type are summarized. The path difference attributes are recorded in groups and missing segments are filled in to generate a path jump structure feature set.

[0019] Based on the path transition structure feature set, a mapping sequence is constructed by encoding all transition nodes and the response interval time. A unified numbering format is set and a unified numbering conversion is performed. The path type, response time period and numbering position are integrated into a structure matrix. The offset value analysis is performed on each transition segment in the matrix to obtain a set of state divergence point locations.

[0020] As a further aspect of the present invention, the specific steps for generating the abnormal state reconstruction path group are as follows:

[0021] Based on the state divergence point map set, a long short-term memory neural network is used to perform time alignment operation between the state change sequence and the current sequence. The interception range is determined by extracting the state start and end index and the corresponding current value interval is extracted. The amplitude change of adjacent data points is calculated and the abrupt change position index is marked. The state synchronous abrupt change position is extracted and an associated index table is constructed to generate the corresponding group of state current abrupt change.

[0022] Based on the state current mutation correspondence group, a one-to-one matching operation between the state and the current mutation point position is performed. By detecting the index difference between the two types of mutation points and setting a threshold to filter out misaligned segments, the unmatched state segments are re-marked and numbered and then separated from the original sequence for storage. All inserted segments are subjected to index offset processing to generate a set of recombined state inserted segments.

[0023] Based on the recombined state insertion fragment set, the original state sequence and the inserted fragment are spliced ​​together. The order deviation of the execution is judged and discontinuous segments are processed by extracting the numbers before and after each splicing point. The merged state stream is sorted according to the triggering order and named with a unified index to obtain the abnormal state reconstruction path group.

[0024] As a further aspect of the present invention, the long short-term memory neural network is configured according to the formula:

[0025]

[0026] in: Indicates time The hidden state vector. This represents the activation function, used for nonlinear transformations. This represents the weight adjustment coefficient of the input path. This represents the weight matrix input to the hidden layer. Indicates time The composite input feature vector, This represents the weight adjustment coefficient of the recursive path. This represents the recursive weight matrix from the previous hidden state to the current state. This represents the hidden state vector from the previous time step. The weighting factor representing the mutation density channel. Indicates time Density of state transitions within the time window The weighting factor representing the stability channel. Indicates time Stability indicators of current signals The bias term representing the input path. The bias term represents the recursive path.

[0027] As a further aspect of the present invention, the Long Short-Term Memory (LSTM) neural network first performs time alignment processing on the extracted state change sequence and the corresponding current data segment to form an input sequence pair. Then, the sequence pair is input into the LSM neural network unit. The transmission and updating of state information are controlled through input gates, forget gates, and output gates, and the data of each time step in the time series is processed step by step. At the same time, the network records historical state information and combines it with the current input to output the predicted state change trend. By comparing the prediction result with the actual state change location, the corresponding time point of the change is extracted and the synchronous change index of state and current is output to construct the state-current change correspondence group.

[0028] As a further aspect of the present invention, the specific steps for generating the jump behavior risk index cluster are as follows:

[0029] Based on the abnormal state, the path group is reconstructed, and a jump segment identification operation is performed. By calculating the difference between consecutive state numbers, non-single-step number pairs are extracted and the start and end point indices are marked. For each group of number jump segments, a corresponding sequence slice is generated and the positional relationship within the sequence is recorded to generate a jump segment index set.

[0030] Based on the skip segment index set, the response residual calculation operation is performed. The response time corresponding to the start and end states of each segment is extracted and the difference is calculated. Then, it is compared with the fixed action tolerance limit. All segments that exceed the tolerance value are marked and associated with the state number and trigger time to generate a list of response abnormal segments.

[0031] Based on the list of abnormal response segments, the abnormal segment device numbers and jump positions are classified. A two-dimensional grouping structure is constructed by extracting the state start number and device number of each segment. Random forest is used to model jump risk with device number, number of abnormal segments, and residual distribution as multiple inputs. The behavior score is output and an abnormal frequency index matrix is ​​generated to obtain the jump behavior risk index cluster.

[0032] As a further aspect of the present invention, the random forest is defined according to the formula:

[0033]

[0034] in: This indicates the jump risk score. This represents the total number of decision trees in the random forest. Indicates the first The weight coefficients of each decision tree. Indicates the first The prediction function of a decision tree. Indicates the device number. Indicates the number of abnormal segments in the corresponding device. This represents the set of residual distribution characteristics of the device. The coefficient of variation represents the state transition cycle. This represents the jump frequency per unit time.

[0035] As a further aspect of the present invention, the random forest first constructs an input sample set characterized by device number, number of abnormal segments, and response residual distribution parameters, forming a multi-dimensional feature vector for each device. Then, multiple decision trees are generated using the training set. Each tree is constructed by random sampling with replacement from the original samples, and nodes are partitioned by randomly selecting some features to generate a structurally differentiated classification model. After all decision trees are constructed, a majority voting mechanism is used to classify the risk of each jump behavior, outputting the jump risk level corresponding to each device and generating an anomaly frequency index matrix.

[0036] As a further aspect of the present invention, the specific steps for generating the risk sequence list of the high-frequency jumping device are as follows:

[0037] Based on the aforementioned jump behavior risk index cluster, the device number extraction operation in the jump segment is performed. By reading the device bit code in the index field and sorting it by segment, the corresponding start time and end time of each jump segment are extracted and converted into a unified time unit. The jump segment identifier index is extracted and bound and mapped with the time point to generate a jump segment device index information table.

[0038] Based on the jump segment device index information table, the number of jump segments under each group of device numbers is counted. The jump frequency table is formed by accumulating the number of segments corresponding to each number. The response duration of each segment is obtained by subtracting the start and end times. After normalizing and transforming the response duration of each segment, it is combined with the jump frequency by column to form a two-dimensional matrix, generating a jump behavior response parameter matrix.

[0039] Based on the jump behavior response parameter matrix, the jump frequency corresponding to each device number is multiplied by the response residual value. The total score is obtained by summarizing the product values ​​according to the device number. The score values ​​are sorted by size and renumbered according to the sequence number to arrange the result table. The sorting results are output in association with the device number to obtain the risk sequence table of high-frequency jump devices.

[0040] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0041] In this invention, the start time and response time in the tripping state transition sequence of the protection device are extracted, the execution delay and action segment length are calculated, and abnormal links are identified based on the state number sequence comparison. This enables the parsing and standardized reconstruction of the state logic structure, forming an ordered tripping path. At the same time, the transition nodes with the same triggering conditions but different state successors are screened, clarifying the deviation relationship between state structure changes and the consistency of triggering conditions, and generating a state transition map with behavioral divergence description capabilities.

[0042] In this invention, a long short-term memory neural network is introduced to time-align the state change sequence with the current signal segment. Through the linkage analysis of the state-current abrupt change position, potential mismatches or abnormal change trends in the state chain are identified, and the state path is reconstructed to enhance the path continuity and dynamic adaptability.

[0043] In this invention, a random forest is used to train and model the device number, number of jump segments, and response residual as multiple inputs. The device score is output through an integrated tree voting mechanism to identify and rank devices with significant abnormal frequency, effectively supporting the system to proactively investigate, prioritize, and rank high-risk devices based on their risk level. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0046] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0047] Example 1

[0048] Please see Figure 1 This invention provides a technical solution: a verification method for an intelligent integrated protection device, comprising the following steps:

[0049] S1: Based on the event change sequence and switch status record recorded during the operation of the protection device, the start time and trip response time are extracted, the execution delay and action duration are calculated, and the status number is compared for sequential consistency to obtain the trip logic transfer path set.

[0050] S2: Based on the set of tripping logic transfer paths, filter the transfer nodes in the interlocking logic module that have the same trigger condition value but different state successors, extract the differences in jump time and path structure, and generate a set of state divergence point locations.

[0051] S3: Based on the state divergence point map, a long short-term memory neural network is used to extract the state change sequence and current data fragments in the alarm triggering logic, perform state sequence comparison and mutation detection, insert reconstruction fragments to form a sequence path flow, and obtain abnormal state reconstruction path groups.

[0052] S4: Based on the abnormal state, reconstruct the path group, identify the position of the jump segment in the regression response mechanism, calculate the response residual and judge the over-limit segment, use random forest to locate the state start point corresponding to the jump segment and classify and summarize it to obtain the jump behavior risk index cluster.

[0053] S5: Based on the risk index cluster of jumping behavior, extract the device number and time information corresponding to the jumping segment, count the response residual and the number of jumping, score and sort, and establish a risk sequence table of high-frequency jumping devices.

[0054] The specific steps for generating the trip logic transfer path set are as follows:

[0055] Based on the event change sequence and switch status record recorded during the operation of the protection device, the time of the first and last nodes in each sequence is read. By performing the difference calculation between the start time and the completion time and uniformly numbering the node labels, the time unit of each record is checked and converted into a unified reference time format. After removing abnormal time records and completing the cleaning, the duration within the tripping section is calculated, and a tripping response time parameter group is generated.

[0056] Based on the trip response time parameter group, the status numbers of each segment are sorted. The index difference of the consecutive number sequence is compared item by item to determine whether it is the minimum single step value. Reverse matching verification is performed. Numbers with gaps are marked and offset compensation is performed. The number stream that passes the consistency check is structured and organized to generate a status number sequence structure set.

[0057] Based on the status number sequence structure set, the number sequence is bound to the time parameter. Each group of numbers corresponds to the response time period and is rearranged into a time stream. The action triggering stage and recovery stage in the time stream are divided. The sequence is spliced ​​and combined in segments and the action chain is reorganized according to the triggering order to obtain the trip logic transfer path set.

[0058] Based on the event change sequence and switch status records recorded during the operation of the protection device, a time field parsing and difference calculation method is used to manually standardize the time information of the first and last nodes in each event record. The time information is read in the format of year-month-day hour:minute:second.millisecond and converted into a unified time unit of milliseconds. The difference between the start time and the end time is calculated. The time span of the state segment is calculated using an integer timestamp. The event tags are mapped through a number table and uniformly numbered. The time fields of records from different sources are checked to the millisecond base format, and the time granularity is unified to 1 millisecond. Records that span days or have reversed time sequences are removed. Records with missing or abnormal time fields are deleted. Then, the start and end time difference corresponding to each jump state is calculated, the duration of each jump segment is calculated, and a trip response time parameter group is generated. The time information in the original event data is standardized and unified, and the data legality is checked, providing a complete time constraint basis for the subsequent state sequence construction.

[0059] Based on the trip response time parameter group, a status number continuity comparison and compensation method is adopted. Each status number sequence in the trip segment is arranged in ascending order. The difference judgment operation between every two adjacent numbers is performed to calculate whether the index difference between the numbers is a single-step increment. The number difference threshold is set to 1, and the trip number points are selected as interruption marks. Then, offset compensation is performed on the records with numbered trip segments. Missing numbers are inserted into the original number stream in sequence and compensation marks are established. Logical consistency checks are performed on the compensated number stream and residual conflict items are removed. A structured number stream is constructed for all number sequences that pass the check, generating a status number sequence structure set and constructing a continuous and missing status number sequence, which helps to accurately map the action stage and improve path consistency.

[0060] Based on the state number sequence structure set, a number-time period correspondence reorganization method is adopted to bind state numbers with trip response time periods one by one, establish the start time and end time period structure corresponding to each state number, rearrange the bound number-time pairs according to the number order, divide them into action triggering stage and state recovery stage, segment them according to the stage boundary labels in the number stream, and splice and combine each stage in sequence to merge into a complete action path chain, remove empty or duplicate segments in the spliced ​​path and re-sort them to obtain the trip logic transfer path set, integrate discrete numbers and time parameters into a complete path stream, realize the reconstructable and serializable expression of the action chain.

[0061] The specific steps for generating the state divergence point atlas are as follows:

[0062] Based on the trip logic transfer path set, the input trigger conditions of all state segments in the interlocking logic module are extracted, the state pairs with completely identical input conditions are grouped according to the input condition values, all subsequent state information of the state pairs are obtained and the types are labeled, the state pairs with structural differences are identified and assigned to the same group set, and a trigger condition consistent branch group is generated.

[0063] Based on the consistent triggering condition branch group, the difference between the start time difference and the response time span of each state segment is calculated. The correspondence between the triggering time and the sequential number of the jump node is matched. The path length between each pair of nodes is calculated and its jump direction and type are summarized. The path difference attributes are recorded in groups and missing segments are filled in to generate a path jump structure feature set.

[0064] Based on the path transition structure feature set, a mapping sequence is constructed by encoding all transition nodes and the response interval time. A unified numbering format is set and a unified numbering conversion is performed. The path type, response time period and numbering position are integrated into a structure matrix. The offset value analysis is performed on each transition segment in the matrix to obtain a set of state divergence point locations.

[0065] Based on the trip logic transfer path set, a condition grouping and state pair matching method is adopted. The input trigger conditions of all state segments in the path set are extracted one by one. The character field comparison method is used to match the values ​​of each input condition field. The judgment rule is that the input fields are completely identical as the filtering criterion. Then, the input fields of each state segment are used as primary keys through the key-value pair mapping mechanism. They are grouped according to the same input value. The state segment numbers of each group are extracted to form state pairs. All subsequent state records in the state pairs are extracted. The subsequent states are classified into adjacent type and skip type according to the number size. The type mark value is set to 1 and 2 for each type of subsequent state, respectively. The marked state pairs are classified and archived by group. The structural difference state groups are summarized by field association method to generate the trigger condition consistent branch group.

[0066] Based on the consistent triggering condition branching group, a path difference calculation and structural difference clustering method is adopted. The difference calculation is performed on the start time and corresponding response time period of each state segment. The time field is set to milliseconds, the start field is start_time, the response field is end_time, and the difference field is duration. The calculation operation is end_time minus start_time. Then, the trigger time of the state segment and the jump node number are paired. The number difference calculation rule is set to the next item minus the previous item. The matching rule is that if the number difference is not equal to 1, it is marked as a jump segment. After extracting the jump segments, the number of jumps between each pair of nodes is counted using the path length calculation method. The path length is counted in absolute value mode. Then, the jump direction is marked forward and backward. Forward is marked as F and backward is marked as B. Finally, the path difference type and direction of each group of state pairs are labeled and aggregated and recorded by number. For missing number segments in discontinuous jump paths, the missing number segments are filled in by interpolation to generate a path jump structure feature set.

[0067] Based on the path transition structure feature set, a number encoding conversion and response mapping construction method is adopted. All transition node numbers are uniformly encoded, with a unified number format of 6-digit integers, padded with zeros if necessary (e.g., original number 23 is converted to 000023). The response time field is calculated in milliseconds, and a correspondence is established between each number and time period, constructing a mapping sequence of number-response interval pairs. The path type field is limited to {adjacent, jump}, and the number position is limited to {first, middle, last}. The path type, response interval, and number position are constructed as triples and uniformly combined into a structure matrix. After sorting by the number field, the matrix rows are traversed. For each transition segment, the absolute difference between the current number position and the previous number position is recorded as the offset value, and the record field is set as offset. The offset values ​​of all transition segments are written into the corresponding structure matrix rows to obtain the state divergence point location map.

[0068] The specific steps for generating abnormal state reconstruction path groups are as follows:

[0069] Based on the state divergence point map, a long short-term memory neural network is used to perform time alignment operation between the state change sequence and the current sequence. The interception range is determined by extracting the state start and end index and the corresponding current value interval is extracted. The amplitude change of adjacent data points is calculated and the abrupt change position index is marked. The state synchronous abrupt change position is extracted and an associated index table is constructed to generate the corresponding group of state current abrupt change.

[0070] Based on the state-current mutation correspondence group, a one-to-one matching operation between the state and the current mutation point position is performed. By detecting the index difference between the two types of mutation points and setting a threshold, misaligned segments are filtered out. The mismatched state segments are re-marked and numbered and then separated from the original sequence. All inserted segments are index offset processed to generate a set of recombined state inserted segments.

[0071] Based on the recombined state insertion fragment set, the original state sequence and the inserted fragment are spliced ​​together. The order deviation is judged and discontinuous segments are processed by extracting the front and back numbers of each splicing point. The merged state stream is sorted according to the triggering order and uniformly named with indexes to obtain the abnormal state reconstruction path group.

[0072] Based on the state divergence point atlas, a long short-term memory neural network is used to perform time alignment operations on the state change sequence and the current sequence. The input format is set as a two-dimensional tensor shape [t, f], where t is the time step and f is the number of features input in each step. The state sequence and the current sequence are constructed as two independent input tensors. The state start index and end index are extracted by a sliding window method. The window length is set to 20 steps and the step size is 5 steps. The original current data range is extracted as input. The input current value is normalized using the min-max scaling rule, and the range is set to [0, 1]. Then, the magnitude change of adjacent data points is calculated in each window by subtracting the previous value from the current value. The data point positions where the change exceeds the set threshold are extracted, and the corresponding index is recorded as the mutation point. The time index of the state change point and the current mutation point is extracted. The mutation position association mapping is established by judging the corresponding position coincidence. An index table containing the state number, current mutation index, and synchronization time field is constructed to generate the state and current mutation corresponding group.

[0073] Based on the corresponding groups of state and current mutations, a mutation point matching and number reconstruction method is adopted. A one-to-one matching operation is performed on the indexes of state mutation points and current mutation points. The index difference threshold is set to ±2. The index difference between each group of state points and current points is calculated. Records with an absolute difference greater than 2 are marked as unaligned segments. The unmatched state number segments are renumbered by adding an offset value of 1000 to the original number for differentiation. The renumbered segments are stored independently from the main sequence. The new number segments are set as a separate insertion segment area. The insertion segments are processed by index offset. The offset value is generated by incrementing in the renumbering order. The insertion number stream is updated by adding 1 from the offset starting point to ensure the numbering is continuous, and a reconstructed state insertion segment set is generated.

[0074] Based on the recombined state insertion fragment set, a state splicing and number sorting method is used to merge the original state sequence and the inserted fragments. The numbers before and after each splicing point are extracted, and a sequence deviation judgment operation is performed. The number order judgment condition is that the difference between the later number and the earlier number is 1. If the condition is not met, the point is marked as a discontinuous segment, its number pair is recorded, and a correction strategy is called. The correction method is to insert virtual number segments to fill the number breakpoints. The virtual number segments are named in the form of "999xxx" and added to the number stream. Then, after merging all numbers, a complete state stream is constructed. The number fields in the merged sequence are reordered in ascending order, and a unified index naming rule is set for each number. The number field is named in the form of "ID_serial number". Finally, the number reconstruction after splicing is completed, and the abnormal state reconstruction path group is obtained.

[0075] Long Short-Term Memory (LSTM) neural networks, according to the formula:

[0076]

[0077] in: Indicates time The hidden state vector. This represents the activation function, used for nonlinear transformations. This represents the weight adjustment coefficient of the input path. This represents the weight matrix input to the hidden layer. Indicates time The composite input feature vector, This represents the weight adjustment coefficient of the recursive path. This represents the recursive weight matrix from the previous hidden state to the current state. This represents the hidden state vector from the previous time step. The weighting factor representing the mutation density channel. Indicates time Density of state transitions within the time window The weighting factor representing the stability channel. Indicates time Stability indicators of current signals The bias term representing the input path. The bias term representing the recursive path;

[0078] The Long Short-Term Memory (LSTM) neural network first aligns the extracted state change sequence with the corresponding current data segment in time to form an input sequence pair. Then, the sequence pair is input into the LSM neural network unit. The input gate, forget gate, and output gate control the transmission and updating of state information, and process the data of each time step in the time series step by step. At the same time, the network records historical state information and combines it with the current input to output the predicted state change trend. By comparing the prediction result with the actual state change position, the corresponding time point of the change is extracted and the synchronous change index of state and current is output to construct the state-current change correspondence group.

[0079] Execution process: First, obtain the composite input feature vector at the current time. It consists of a status code, current value, current difference, and direction of change, representing the multidimensional input signal of the device under the current operating condition. Then, The hidden state vector from the previous time step The data is fed into the network through the input path and the recursive path, respectively, and the weight coefficients for each path are as follows: and This indicates the degree to which the model depends on the current input features and historical states. It is obtained automatically during the training phase by minimizing the error. Then, the current mutation density parameter is introduced. This represents the number of abrupt changes per unit time within the current time window, reflecting the frequency of fluctuations in the current system state. A current stability parameter is also introduced. Defined as the ratio of the standard deviation to the average value of the current within the current time window, it characterizes the fluctuation amplitude of the current signal. Both are multiplied by the channel weighting coefficient. and The calculation path is then incorporated, and the model's discrimination accuracy is ensured through dynamic adjustments using information gain scoring and loss function regularization. Finally, all weighted terms and bias terms are summed and passed through an activation function. Perform a nonlinear mapping and output the hidden state vector at the current time step. It is used to support subsequent state-current synchronous change detection and verification judgment, and to realize synchronous consistency verification of the state logic and physical electrical quantities of the protection device.

[0080] The specific steps for generating a risk index cluster for jump behavior are as follows:

[0081] Based on the abnormal state, the path group is reconstructed, and the jump segment identification operation is performed. The non-single-step number pairs are extracted by calculating the difference of the consecutive state numbers and the start and end point indices are marked. For each group of number jump segments, the corresponding sequence slices are generated and the positional relationship within the sequence is recorded to generate a jump segment index set.

[0082] Based on the skip segment index set, the response residual calculation operation is performed. By extracting the response time corresponding to the start and end states of each segment and calculating the difference, it is then compared with the fixed action tolerance limit. All segments that exceed the tolerance value are marked and associated with the state number and trigger time to generate a list of response abnormal segments.

[0083] Based on the list of response anomaly segments, the device numbers and jump positions of the anomaly segments are classified. A two-dimensional grouping structure is constructed by extracting the state start number and device number of each segment. Random forest is used to model the jump risk with device number, number of anomaly segments, and residual distribution as multiple inputs. The behavior score is output and an anomaly frequency index matrix is ​​generated to obtain the jump behavior risk index cluster.

[0084] Based on the abnormal state reconstruction path group, the state number difference filtering method is adopted. The difference between the numbers is calculated for all state numbers. The difference calculation method is set to the current number minus the previous number. The single step threshold is set to 1. Number pairs with a difference of not equal to 1 are filtered and their index positions are recorded. All discontinuous number segments are extracted and the start and end numbers are marked. According to the recorded index start and end values, the corresponding segments in the original state stream are extracted. Each segment is constructed as a group of state number sequence slices and recorded. The offset format is the distance from the start number position to the end number position. Finally, all jump segments are sorted out according to the segment order and output as number jump segment index group, generating jump segment index set.

[0085] Based on the skip segment index set, the response time residual comparison method is adopted to extract the corresponding response time field for each start and end state in the skip segment. The format is uniformly set as millisecond timestamp. The calculation method is to subtract the start time from the end time. The calculation result is defined as the response time difference residual. The tolerance threshold is set to 500 milliseconds. Segments with a time difference residual greater than the threshold are used for filtering. The filtering rule is that if the time difference residual is greater than 500, it is judged as an out-of-limit segment. For all out-of-limit segments, the start state number, residual value and corresponding trigger time are recorded. All out-of-limit segments are summarized in order of number and output as a time abnormal segment record list to generate a response abnormal segment list.

[0086] Based on the list of response anomaly segments, a random forest algorithm is used. The device number, number of anomaly segments, and response residual value in each anomaly record are used as feature inputs to construct an input feature vector in the format [X1, X2, X3]. X1 is the value corresponding to the device number, X2 is the number of anomaly segments corresponding to the number, and X3 is the average of all residual values. The training set group ratio is set to 7:3. The random forest classification method is used to perform modeling operations on the training data. The number of trees is set to 100, the maximum depth is 10, and information gain is used as the splitting criterion. A random forest model is generated through training. The model outputs the anomaly score value corresponding to each device. Then, the score values ​​are sorted, with the sorting field set to score values ​​from high to low. Finally, all scores are summarized to form a two-dimensional score matrix with the device number and jump frequency to obtain the jump behavior risk index cluster.

[0087] Random forest, according to the formula:

[0088]

[0089] in: This indicates the jump risk score. This represents the total number of decision trees in the random forest. Indicates the first The weight coefficients of each decision tree. Indicates the first The prediction function of a decision tree. Indicates the device number. Indicates the number of abnormal segments in the corresponding device. This represents the set of residual distribution characteristics of the device. The coefficient of variation represents the state transition cycle. This indicates the jump frequency per unit time.

[0090] Random Forest first constructs an input sample set featuring device number, number of anomalous segments, and response residual distribution parameters. For each device, a multi-dimensional feature vector is formed. Then, multiple decision trees are generated using the training set. Each tree is constructed by random sampling with replacement from the original samples, and nodes are split by randomly selecting some features to generate a structurally differentiated classification model. After all decision trees are constructed, a majority voting mechanism is used to classify the risk of each jump behavior, output the jump risk level corresponding to each device, and generate an anomaly frequency index matrix.

[0091] Execution process: First, extract the serial number of each device. Count the number of corresponding abnormal segments. Then, based on historical behavior data, the residual distribution characteristics are calculated. This includes the maximum deviation, mean deviation, standard deviation, and skewness between observed values ​​and the theoretical model within the anomaly segment, used to measure the stability of the response anomaly. Next, the state transition time series is extracted, and the periodic variation coefficient is calculated. This serves as a quantitative indicator of whether the state transition rhythm fluctuates abnormally, and also calculates the jump frequency. This refers to the number of discontinuous state transitions occurring per unit time, used to measure the dynamic instability of the device. The input features form a feature vector, which is then fed into each decision tree. Risk prediction is performed using weighting coefficients. The predicted outputs for each tree are weighted. The prediction accuracy of each tree on the training set is normalized to determine the final result. The weighted outputs of the trees are summed to obtain the jump risk score. It is used to establish an anomaly frequency index matrix and divide the jump behavior risk index cluster.

[0092] The specific steps for generating the risk sequence list of high-frequency jumping devices are as follows:

[0093] Based on the risk index cluster of skipping behavior, the device number extraction operation in the skipping segment is performed. By reading the device bit code of the index field and sorting it by segment, the corresponding start time and end time of each skipping segment are extracted and converted into a unified time unit. The skipping segment identifier index is extracted and bound and mapped with the time point to generate a skipping segment device index information table.

[0094] Based on the jump segment device index information table, the number of jump segments under each group of device numbers is counted. The jump frequency table is formed by accumulating the number of segments corresponding to each number. The response duration of each segment is obtained by subtracting the start and end times. After normalizing and transforming the response duration of each segment, it is combined with the jump frequency by column to form a two-dimensional matrix, generating a jump behavior response parameter matrix.

[0095] Based on the jump behavior response parameter matrix, the jump frequency corresponding to each device number is multiplied with the response residual value. The total score is obtained by summarizing the product values ​​by device number. The score values ​​are sorted by size and the result table is renumbered by sequence number. The sorted results are output in association by device number to obtain the risk sequence table of high-frequency jump devices.

[0096] Based on the risk index cluster of skip behavior, a field parsing and time standardization method is used to extract the device number field of the records in the skip segment. The device code marked in each record is used as the grouping basis for classification and organization. The start time and end time of each group of skip segment records are extracted and uniformly read in the format of year-month-day hour:minute:second.millisecond, then transcribed into a uniform pure number format and recorded in milliseconds. The skip identification number of each skip segment information is extracted according to the original index field and a correspondence between it and the time field is established. Finally, the skip segment index, device number, and start and end time are encapsulated into a field structure to generate a skip segment device index information table. This completes the standardized extraction of device numbers and uniform time binding of the skip segment structure, forming a statistical data infrastructure.

[0097] Based on the skip segment device index information table, the method of counting and response time normalization is adopted to count the skip segment entries under each device number group, and accumulate the skip count results corresponding to each device number. At the same time, the start time and end time in each skip segment record are directly subtracted to obtain the response duration. All response duration values ​​are uniformly linearly mapped and compressed to a normalized value between 0 and 1. The normalized response duration and skip count are merged into a two-field record of the same device by row matching. Each record is output as a three-element table entry containing device number, skip count, and normalized response duration, generating a skip behavior response parameter matrix, clarifying the frequency and persistence characteristics of each device in response behavior, and constructing the basis for subsequent scoring input.

[0098] Based on the jump behavior response parameter matrix, a behavior scoring and hierarchical sorting method is adopted. The jump frequency value and response time normalization value corresponding to the device number in each record are multiplied to obtain the corresponding behavior score value. Then, the devices are grouped by device number, and all scores are summed to obtain the total score result for each device number. The score values ​​are sorted and renumbered according to the order from high to low to generate a sorting field. The sorting field is then paired with the corresponding device number and sorted. Finally, the device number, total score value and sorting sequence number of each record are output to obtain a risk sequence list of high-frequency jump devices. A device hierarchical system based on the quantitative superposition of abnormal features is constructed to form an orderly risk identification list.

[0099] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A verification method for an intelligent integrated protection device, characterized in that, Includes the following steps: S1: Based on the event change sequence and switch status record recorded during the operation of the protection device, the start time and trip response time are extracted, the execution delay and action duration are calculated, and the status number is compared for sequential consistency to obtain the trip logic transfer path set. S2: Based on the trip logic transfer path set, filter the transfer nodes in the interlocking logic module that have the same trigger condition value but different state successors, extract the differences in jump time and path structure, and generate a state divergence point map set. S3: Based on the state divergence point map set, a long short-term memory neural network is used to extract the state change sequence and current data fragment in the alarm triggering logic, perform state sequence comparison and mutation detection, insert reconstruction fragments to form a sequence path flow, and obtain abnormal state reconstruction path group; S4: Based on the abnormal state, reconstruct the path group, identify the position of the jump segment in the regression response mechanism, calculate the response residual and judge the over-limit segment, use random forest to locate the state starting point corresponding to the jump segment and classify and summarize it to obtain the jump behavior risk index cluster. S5: Based on the jump behavior risk index cluster, extract the device number and time information corresponding to the jump segment, count the response residual and the number of jumps, score and sort, and establish a risk sequence table of high-frequency jump devices. The Long Short-Term Memory neural network is configured according to the formula: in: Indicates time The hidden state vector. This represents the activation function, used for nonlinear transformations. This represents the weight adjustment coefficient of the input path. This represents the weight matrix input to the hidden layer. Indicates time The composite input feature vector, This represents the weight adjustment coefficient of the recursive path. This represents the recursive weight matrix from the previous hidden state to the current state. This represents the hidden state vector from the previous time step. The weighting factor representing the mutation density channel. Indicates time Density of state transitions within the time window The weighting factor representing the stability channel. Indicates time Stability indicators of current signals The bias term representing the input path. The bias term represents the recursive path.

2. The verification method for the intelligent integrated protection device according to claim 1, characterized in that, The specific steps for generating the trip logic transfer path set are as follows: Based on the event change sequence and switch status record recorded during the operation of the protection device, the time of the first and last nodes in each sequence is read. By performing the difference calculation between the start time and the completion time and uniformly numbering the node labels, the time unit of each record is checked and converted into a unified reference time format. After removing abnormal time records and completing the cleaning, the duration within the tripping section is calculated, and a tripping response time parameter group is generated. Based on the trip response time parameter group, the status numbers of each segment are sorted. The index difference of the consecutive number sequence is compared item by item to determine whether it is the minimum single step value. Reverse matching verification is performed. Numbers with gaps are marked and offset compensation is performed. The number stream that passes the consistency check is structured and organized to generate a status number sequence structure set. Based on the state number sequence structure set, the number sequence is bound to the time parameter, each group of numbers is sequentially corresponding to the response time period and rearranged into a time stream, the action triggering stage and the recovery stage in the time stream are divided, the sequence is spliced ​​and combined in segments and the action chain is reorganized according to the triggering order to obtain the trip logic transfer path set.

3. The verification method for the intelligent integrated protection device according to claim 1, characterized in that, The specific steps for generating the state divergence point map set are as follows: Based on the trip logic transfer path set, the input trigger conditions of all state segments in the interlocking logic module are extracted, grouped according to the input condition values, and state pairs with identical input values ​​are selected. All subsequent state information of the state pairs are obtained and type-labeled. State pairs with structural differences are identified and grouped into the same set to generate a trigger condition consistent branch group. Based on the consistent branching group of the triggering conditions, the difference between the start time difference and the response time span of each state segment is calculated. The correspondence between the triggering time and the sequential number of the jump node is matched. The path length between each pair of nodes is calculated and their jump direction and type are summarized. The path difference attributes are recorded in groups and missing segments are filled in to generate a path jump structure feature set. Based on the path transition structure feature set, a mapping sequence is constructed by encoding all transition nodes and the response interval time. A unified numbering format is set and a unified numbering conversion is performed. The path type, response time period and numbering position are integrated into a structure matrix. The offset value analysis is performed on each transition segment in the matrix to obtain a set of state divergence point locations.

4. The verification method for the intelligent integrated protection device according to claim 1, characterized in that, The specific steps for generating the abnormal state reconstruction path group are as follows: Based on the state divergence point map set, a long short-term memory neural network is used to perform time alignment operation between the state change sequence and the current sequence. The interception range is determined by extracting the state start and end index and the corresponding current value interval is extracted. The amplitude change of adjacent data points is calculated and the abrupt change position index is marked. The state synchronous abrupt change position is extracted and an associated index table is constructed to generate the corresponding group of state current abrupt change. Based on the state current mutation correspondence group, a one-to-one matching operation between the state and the current mutation point position is performed. By detecting the index difference between the two types of mutation points and setting a threshold to filter out misaligned segments, the unmatched state segments are re-marked and numbered and then separated from the original sequence for storage. All inserted segments are subjected to index offset processing to generate a set of recombined state inserted segments. Based on the recombined state insertion fragment set, the original state sequence and the inserted fragment are spliced ​​together. The order deviation of the execution is judged and discontinuous segments are processed by extracting the numbers before and after each splicing point. The merged state stream is sorted according to the triggering order and named with a unified index to obtain the abnormal state reconstruction path group.

5. The verification method for the intelligent integrated protection device according to claim 4, characterized in that, The Long Short-Term Memory (LSTM) neural network first performs time alignment processing on the extracted state change sequence and the corresponding current data segment to form an input sequence pair. Then, the sequence pair is input into the LSM neural network unit. The transmission and updating of state information are controlled through input gates, forget gates, and output gates, and the data of each time step in the time series is processed step by step. At the same time, the network records historical state information and combines it with the current input to output the predicted state change trend. By comparing the prediction result with the actual state change position, the corresponding time point of the change is extracted and the synchronous change index of state and current is output to construct the state-current change correspondence group.

6. The verification method for the intelligent integrated protection device according to claim 1, characterized in that, The specific steps for generating the aforementioned jump behavior risk index cluster are as follows: Based on the abnormal state, the path group is reconstructed, and a jump segment identification operation is performed. By calculating the difference between consecutive state numbers, non-single-step number pairs are extracted and the start and end point indices are marked. For each group of number jump segments, a corresponding sequence slice is generated and the positional relationship within the sequence is recorded to generate a jump segment index set. Based on the skip segment index set, the response residual calculation operation is performed. The response time corresponding to the start and end states of each segment is extracted and the difference is calculated. Then, it is compared with the fixed action tolerance limit. All segments that exceed the tolerance value are marked and associated with the state number and trigger time to generate a list of response abnormal segments. Based on the list of abnormal response segments, the abnormal segment device numbers and jump positions are classified. A two-dimensional grouping structure is constructed by extracting the state start number and device number of each segment. Random forest is used to model jump risk with device number, number of abnormal segments, and residual distribution as multiple inputs. The behavior score is output and an abnormal frequency index matrix is ​​generated to obtain the jump behavior risk index cluster.

7. The verification method for the intelligent integrated protection device according to claim 6, characterized in that, The random forest is defined according to the formula: in: This indicates the jump risk score. This represents the total number of decision trees in the random forest. Indicates the first The weight coefficients of each decision tree. Indicates the first The prediction function of a decision tree. Indicates the device number. Indicates the number of abnormal segments in the corresponding device. This represents the set of residual distribution characteristics of the device. The coefficient of variation represents the state transition cycle. This represents the jump frequency per unit time.

8. The verification method for the intelligent integrated protection device according to claim 6, characterized in that, The random forest first constructs an input sample set characterized by device number, number of anomalous segments, and response residual distribution parameters, forming a multi-dimensional feature vector for each device. Then, multiple decision trees are generated using the training set. Each tree is constructed by random sampling with replacement from the original samples, and nodes are partitioned by randomly selecting some features to generate a structurally differentiated classification model. After all decision trees are constructed, a majority voting mechanism is used to classify the risk of each jump behavior, outputting the jump risk level corresponding to each device and generating an anomaly frequency index matrix.

9. The verification method for the intelligent integrated protection device according to claim 1, characterized in that, The specific steps for generating the risk sequence list of the high-frequency jumping device are as follows: Based on the aforementioned jump behavior risk index cluster, the device number extraction operation in the jump segment is performed. By reading the device bit code in the index field and sorting it by segment, the corresponding start time and end time of each jump segment are extracted and converted into a unified time unit. The jump segment identifier index is extracted and bound and mapped with the time point to generate a jump segment device index information table. Based on the jump segment device index information table, the number of jump segments under each group of device numbers is counted. The jump frequency table is formed by accumulating the number of segments corresponding to each number. The response duration of each segment is obtained by subtracting the start and end times. After normalizing and transforming the response duration of each segment, it is combined with the jump frequency by column to form a two-dimensional matrix, generating a jump behavior response parameter matrix. Based on the jump behavior response parameter matrix, the jump frequency corresponding to each device number is multiplied by the response residual value. The total score is obtained by summarizing the product values ​​according to the device number. The score values ​​are sorted by size and renumbered according to the sequence number to arrange the result table. The sorting results are output in association with the device number to obtain the risk sequence table of high-frequency jump devices.

Images