Residual current real-time remote protection method based on edge calculation

Abstract

The invention discloses a residual current real-time remote protection method based on edge calculation, which relates to the technical field of power failure protection, and comprises the following steps: executing time sorting and differential calculation based on residual current signal data acquired by current sensors deployed on power distribution branches and corresponding acquisition timestamps; generating a residual current signal record containing the time interval value and the amplitude difference value; extracting a signal arrival delay value and a time jitter value of each record according to the residual current signal record and a locally stored model version number, and fitting the two types of data to generate a time drift compensation parameter; performing version consistency comparison based on the time drift compensation parameter and a cloud model version number returned by the cloud interface; according to the method, a synchronous verification and arbitration updating mechanism of the local and cloud residual current anomaly judgment result is constructed in an edge computing environment, so that the consistency of the judgment result can still be kept under the conditions that the network is unstable and the model is frequently updated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power failure protection, in particular to a residual current real-time remote protection method based on edge computing. BACKGROUND

[0002] In the operation process of the power distribution network, the residual current, as an important electrical parameter reflecting the line insulation condition and the risk of electric leakage, is usually continuously collected by the current sensors deployed on each distribution branch, and combined with the electrical safety protection device to perform protection actions such as abnormal power-off.

[0003] With the development of edge computing technology, some power distribution systems begin to configure computing units on the edge nodes close to the power terminal, so that they have the ability to analyze and determine the residual current signal locally, in order to ensure the basic electrical safety protection response capability in the case of network communication delay or high cloud computing load. This method can avoid the communication bottleneck caused by relying on uploading the cloud for reasoning, and to a certain extent, shorten the fault response time and improve the operation safety of the power distribution system.

[0004] However, the existing methods are mainly one-sided determination, usually only relying on the edge node to complete the abnormal determination independently, although it reduces the dependence on the network, but due to the limited resources of the edge node, the model update frequency is low, and it is easy to produce determination deviation with the cloud model after a long time of operation, once the model version lags behind, the recognition accuracy of abnormal current will gradually decrease; At the same time, the existing centralized cloud method can ensure the algorithm accuracy through unified model updating, but when the network is unstable or the transmission delay is too large, it will cause the determination result to be unable to return in time, delay the protection action, and there is a security risk, and there is no verification mechanism for the consistency of the determination results between nodes. SUMMARY

[0005] In order to solve the above technical problems, the present application provides a residual current real-time remote protection method based on edge computing, which comprises: S11, based on the residual current signal data collected by the current sensors deployed on each distribution branch and the corresponding collection time stamp, performing time sorting and difference calculation to generate residual current signal records containing time interval value and amplitude difference value; S12, according to the residual current signal record and the locally stored model version number, extracting the signal arrival delay value and time jitter value of each record, and fitting the two types of data to generate time drift compensation parameters; S13, based on the time drift compensation parameters and the cloud model version number returned by the cloud interface, performing version consistency comparison to generate version consistency confidence value; at the same time, based on the residual current signal record and the local model reasoning to generate local abnormal determination result; S14, triggering the double-judgment process based on the local abnormality judgment result and the version consistency confidence value, calling the cloud model reasoning to obtain the cloud abnormality judgment result, and comparing the local abnormality judgment result with the cloud abnormality judgment result for consistency to generate a conflict marker record; S15, performing arbitration fusion operation based on the conflict marker record and the corresponding residual current signal record to generate an arbitration judgment result, and calling the arbitration judgment result to perform local model parameter rollback update when the model version number and the cloud model version number are inconsistent.

[0006] Further, the step of generating the residual current signal record is: S111, synchronously acquiring residual current signal data and collection time stamps through current sensors deployed on each power distribution branch; S112, performing time sorting on the residual current signal data based on the collection time stamps to obtain sorted signal data; S113, performing difference calculation according to the difference between adjacent data points of the sorted signal data to obtain amplitude difference value and time interval value; S114, structuring the amplitude difference value and the time interval value to output the residual current signal record.

[0007] Further, the logic for calculating the amplitude difference value and the time interval value is: S113.1, extracting the amplitude and time value of the adjacent two data points in the sorted signal data; S113.2, subtracting the amplitude of the previous data point from the amplitude of the next data point to generate the amplitude difference value; S113.3, subtracting the time value of the previous data point from the time value of the next data point to generate the time interval value.

[0008] Further, the step of generating the time drift compensation parameter is: S121, extracting the start and end time stamps of each record in the residual current signal record to calculate the signal arrival duration to obtain the signal arrival delay value; S122, calculating the time jitter value according to the time interval difference of adjacent records in the residual current signal record; S123, fitting the signal arrival delay value and the time jitter value into the time drift model to obtain the time drift compensation parameter.

[0009] Further, the logic for obtaining the signal arrival delay value is: S121.1, obtaining the start time stamp and the end time stamp of each record in the residual current signal record; S121.2, subtracting the start time stamp from the end time stamp to obtain the signal arrival delay value.

[0010] Further, the step of generating the version consistency confidence value is: S131, acquiring a cloud model version number through a cloud interface and reading a locally stored model version number; S132, performing string comparison and verification based on the model version number and the cloud model version number; S133, mapping the comparison result to a consistency level value and outputting the consistency level value as the version consistency confidence value.

[0011] Further, the step of generating the local anomaly determination result is: S134, inputting the residual current signal record into the local model to obtain an anomaly score value through inference; S135, performing classification operation based on the anomaly score value and a determination threshold set in the model, and marking the classification result as the local anomaly determination result.

[0012] Further, the step of generating the conflict marker record is: S141, synchronously triggering a cloud model inference process based on the local anomaly determination result and the version consistency confidence value; S142, inputting the residual current signal record into the cloud model to obtain a cloud anomaly determination result; S143, comparing the local anomaly determination result and the cloud anomaly determination result item by item and recording the inconsistent items, and outputting the inconsistent items as the conflict marker record.

[0013] Further, the step of calling the arbitration determination result to perform local model parameter backtracking update is: S151, locating the corresponding residual current signal record based on the conflict marker record; S152, inputting the conflict marker record and the residual current signal record into an arbitration fusion operation module to obtain an arbitration fusion output result; S153, outputting the arbitration fusion output result as the arbitration determination result; S154, when the model version number and the cloud model version number are inconsistent, performing local model parameter backtracking update through the arbitration determination result.

[0014] Compared with the prior art, the present application has the following advantages: The present application can continuously monitor and locally identify the residual current state of each power distribution branch through the cooperative processing of the residual current signal record and the local anomaly determination result, so as to realize rapid perception and classification marking of the abnormal current state, thereby improving the immediate response capability of the local to abnormal risks, and enhancing the independent protection capability of the edge side electrical safety; In addition, the application can compare the version states of the local model and the cloud model in real time through joint analysis of the time drift compensation parameter and the version consistency confidence value, so as to realize dynamic evaluation and synchronous correction of the model reasoning reliability, and further improve the accuracy of the determination result in the cross-version environment, thereby improving the stability of the distributed model determination. Further, the application can complete the consistency arbitration of the abnormal label and the backtracking update of the local model parameter in the case that the local and cloud determination results are inconsistent through the closed-loop fusion of the conflict marker record and the arbitration determination result, so as to realize the self-adaptive correction ability of the local model in the continuous operation, and further improve the determination convergence of the model in the long-term operation condition, thereby guaranteeing the continuous reliability of the overall residual current real-time remote protection process. To sum up, the application realizes the consistency of the determination result under the condition of unstable network and frequent model update through the synchronous verification and arbitration update mechanism of the local and cloud residual current abnormal determination result in the edge computing environment, so as to improve the accuracy and continuity of the residual current real-time protection. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art based on these drawings.

[0016] Figure 1 A flowchart of a residual current real-time remote protection method based on edge computing provided for the embodiment 1 of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. Embodiment 1

[0018] Please refer to Figure 1 As shown in the figure, the embodiment discloses a residual current real-time remote protection method based on edge computing, which comprises the following steps: S11, based on the residual current signal data collected by the current sensor deployed on each power distribution branch and the corresponding collection time stamp, performing time sorting and difference calculation to generate residual current signal records containing time interval values and amplitude difference values; Specifically, the step of generating residual current signal records is: S111, synchronously acquiring residual current signal data and collection time stamps by the current sensor deployed on each power distribution branch; In one specific embodiment, the current sensor deployed on each power distribution branch continuously samples the current of each branch conductor to obtain residual current signal data; at the same time, the high-precision clock module inside the current sensor is called to record the starting time of sampling to obtain the corresponding collection time stamp; It should be noted that: the current sensor is an alternating current differential residual current transformer, which can synchronously output sampled current and time stamp with millisecond resolution; the residual current signal data is the instantaneous current amplitude sequence continuously output by the sensor in the sampling period, and the collection time stamp is the absolute time corresponding to each group of sampling amplitudes.

[0019] S112, performing time sorting on the residual current signal data based on the collection time stamp to obtain sorted signal data; In one specific embodiment, the collection time stamp corresponding to each group of residual current signal data is extracted, and the residual current signal data is rearranged in the order of time from early to late to obtain sorted signal data; It should be noted that: the sorted signal data retains the pairing correspondence between the amplitude of each sampling point in the original residual current signal data and the corresponding collection time stamp, only adjusts the time sequence of each sampling point, and does not change the original amplitude content.

[0020] S113, performing difference calculation according to the difference between adjacent data points of the sorted signal data to obtain amplitude difference values and time interval values; In one specific embodiment, two adjacent groups of amplitude and time values in the sorted signal data are selected in turn, the amplitude difference value is calculated by subtracting the amplitude of the previous group from the amplitude of the next group, and the time interval value is calculated by subtracting the time value of the previous group from the time value of the next group; It should be noted that: the amplitude difference value is used to represent the degree of change of the current amplitude of adjacent sampling points, and the time interval value is used to represent the sampling time span between adjacent sampling points.

[0021] Specifically, the logic for calculating amplitude difference values and time interval values is: S113.1, extracting the amplitude and time value of the adjacent two data points in the sorted signal data; S113.2, subtracting the amplitude of the previous data point from the amplitude of the next data point to generate the amplitude difference value; S113.3, subtract the previous data point time value from the latter data point time value to generate a time interval value.

[0022] S114, structurally record the amplitude difference value and the time interval value, and output a residual current signal record.

[0023] In a specific embodiment, the amplitude difference value and the time interval value are combined in pairs according to the sampling order to generate a residual current signal record containing two types of numerical pairs, and the time sequence of each numerical pair is identified by a record index. It should be noted that the residual current signal record is a set of time-ordered data pair records, each record containing the amplitude difference value and the time interval value of the corresponding adjacent sampling points.

[0024] S12, according to the residual current signal record and the locally stored model version number, extracting the signal arrival delay value and the time jitter value of each record, and fitting the two types of data to generate a time drift compensation parameter; Specifically, the step of generating the time drift compensation parameter is: S121, extracting the start and end time stamps of each record in the residual current signal record to calculate the signal arrival duration, and obtaining the signal arrival delay value; In a specific embodiment, the start and end positions of each record in the residual current signal record are located, the respective start time stamp and end time stamp are read, and the signal arrival delay value is calculated by the difference between the two. It should be noted that the signal arrival delay value is used to represent the time length of the propagation of the current signal between the start and end points of sampling, and provides a reference for subsequent evaluation of signal arrival deviation.

[0025] S122, calculating the time jitter value according to the time interval difference of adjacent records in the residual current signal record; In a specific embodiment, the time interval values of adjacent two records in the residual current signal record are extracted in sequence, the latter time interval value is subtracted from the former time interval value, and the time jitter value is obtained; It should be noted that the time jitter value is used to represent the stability of the sampling period between consecutive records, and the larger the value represents the more obvious the fluctuation of the sampling period.

[0026] S123, fitting the signal arrival delay value and the time jitter value to the time drift model to obtain the time drift compensation parameter.

[0027] In a specific embodiment, the signal arrival delay value and the time jitter value are input into a preset time drift model for joint fitting, and the time drift compensation parameter is output; It should be noted that the time drift model is established using a linear regression method based on the mapping relationship between delay fluctuations and jitter deviations in historical sampling records.

[0028] Specifically, the logic for obtaining the signal arrival delay value is as follows: S121.1, Obtain the start and end timestamps of each record in the residual current signal record; In one specific embodiment, the start and end timestamps of all records in the residual current signal record are read and matched according to the record index; It should be noted that the start timestamp and end timestamp are absolute time stamps generated synchronously by the current sensor when it outputs the remaining current signal data for the first and last time within the acquisition period.

[0029] S121.2, subtract the start timestamp from the end timestamp to obtain the signal arrival delay value.

[0030] In one specific embodiment, the corresponding signal arrival delay value is obtained by subtracting the corresponding start timestamp from the end timestamp of each record, and is identified by the record index number. It should be noted that the signal arrival delay value is a time unit value used to represent the total propagation time of the current signal in each record.

[0031] S13, perform version consistency comparison based on time drift compensation parameters and cloud model version number returned by cloud interface to generate version consistency confidence value; at the same time, generate local anomaly judgment result based on residual current signal record and local model inference. Specifically, the steps for generating version consistency confidence values ​​are as follows: S131, obtain the model version number from the cloud via the cloud interface and read the model version number stored locally; In one specific embodiment, a version request instruction is sent to the cloud model management server through the cloud communication interface to obtain the cloud model version number, and the model version number is synchronously read from the local non-volatile memory. It should be noted that: the cloud model version number is the version number of the inference parameters currently loaded in the cloud model, and the model version number is the version number of the parameters currently running in the local model. Both are identified by unique numbers for easy comparison.

[0032] S132, based on the model version number and the cloud model version number, performs string comparison and verification; In one specific embodiment, the obtained model version number is compared with the cloud model version number character by character. If the characters are inconsistent, the index of the difference is recorded and counted. Then, the verification rules are executed to calculate the difference ratio. It should be noted that the difference ratio is obtained by dividing the number of differing bits by the total number of bits, and is used to reflect the degree of deviation between the local and cloud version numbers.

[0033] S133 maps the comparison results to consistency level values ​​and outputs version consistency confidence values.

[0034] In a specific embodiment, based on the difference ratio obtained in S132, the corresponding level number is matched in the preset version consistency level table, and the level number is converted into a numerical version consistency confidence value. It should be noted that the version consistency confidence value ranges from 0 to 1. The closer the value is to 1, the more consistent the model version number is with the cloud model version number.

[0035] Specifically, the steps for generating local anomaly detection results are as follows: S134, The residual current signal is recorded and input into the local model for inference to obtain the anomaly score value; In one specific embodiment, the residual current signal records are sequentially input into the inference interface of the local model according to the record index, and feature parsing and classification network forward inference are performed to obtain the anomaly score value corresponding to each record. It should be noted that the anomaly score is a numerical value of the probability of an anomaly output by the local model based on the weights trained by the historical annotations. It is used to quantify the degree of anomaly in the residual current signal record. The local model is a classification model trained on the historical residual current signal record and its manually annotated anomaly labels, and can output an anomaly score in the range of 0 to 1.

[0036] S135: Perform classification operation based on the anomaly score and the model's built-in judgment threshold, and mark the classification result as the local anomaly judgment result.

[0037] In one specific embodiment, the anomaly score of each record is compared with the built-in judgment threshold of the model. If the anomaly score is greater than the built-in judgment threshold of the model, the "abnormal" label is output; otherwise, the "normal" label is output. The obtained label is used as the local anomaly judgment result. It should be noted that the built-in decision threshold is a fixed value selected through cross-validation accuracy optimization during the model training phase to ensure the consistency and repeatability of local anomaly detection results. The built-in decision threshold is a fixed threshold determined through the optimal F1 score of cross-validation to distinguish between abnormal and normal records.

[0038] S14, based on the local anomaly judgment result and the version consistency confidence value, a dual judgment process is triggered, the cloud model inference is called to obtain the cloud anomaly judgment result, and the consistency comparison between the local anomaly judgment result and the cloud anomaly judgment result is performed to generate a conflict marker record. Specifically, the steps for generating conflict marker records are as follows: S141, based on the local anomaly judgment result and the version consistency confidence value, the cloud model inference process is triggered synchronously. In a specific embodiment, when the version consistency confidence value is detected to be lower than the preset synchronization threshold and the local anomaly determination result of the corresponding record is abnormal, an inference trigger command is immediately generated and sent to the cloud model interface to synchronously start the cloud inference process. It should be noted that the preset synchronization threshold is a fixed value determined based on the statistics of historical version update frequency. It is used to control the triggering frequency of the cloud model inference process and ensure that dual judgment is only initiated when there is a risk of version difference.

[0039] S142, input the residual current signal record into the cloud model to obtain the cloud anomaly judgment result; The corresponding residual current signal records are uploaded to the cloud model interface in chronological order, feature decoding and classification reasoning are performed, and the cloud anomaly judgment results corresponding to the records are output. It should be noted that the cloud-based anomaly determination result is an anomaly determination label calculated based on the inference weights of the cloud model under the current cloud model version number.

[0040] S143 compares the local anomaly determination results with the cloud anomaly determination results one by one and records the inconsistencies, outputting conflict marker records.

[0041] The local anomaly determination results are compared with the corresponding cloud anomaly determination results for each record. All inconsistent records are labeled with conflict tags, and the record numbers with conflict tags are output as conflict marked records. It should be noted that: conflict marker records are used to indicate the record numbers where there are differences between the local anomaly determination results and the cloud anomaly determination results.

[0042] S15: Based on the conflict marker record and the corresponding residual current signal record, perform arbitration fusion operation to generate arbitration judgment result, and when the model version number is inconsistent with the cloud model version number, call the arbitration judgment result to perform local model parameter backtracking update.

[0043] Specifically, the steps for performing local model parameter backtracking updates by invoking the arbitration ruling result are as follows: S151, locate the corresponding residual current signal record based on the collision marker record; In one specific embodiment, each number recorded in the conflict tag record is retrieved, and the record item corresponding to the number is extracted from the residual current signal record to obtain the residual current signal record with conflict tag. It should be noted that the extracted residual current signal records with conflict tags are used as one of the inputs to the arbitration fusion operation to ensure that arbitration only applies to records with conflicting determinations.

[0044] S152, input the conflict mark record and the residual current signal record into the arbitration fusion calculation module to obtain the arbitration fusion output result; Extract the residual current signal records corresponding to the record numbers marked as conflicting in the conflict marking record one by one, read the local anomaly judgment result and the cloud anomaly judgment result of each record respectively, and directly use the tag as the arbitration fusion output when the two are consistent; when the two are inconsistent, manually confirm and mark the record according to the cloud anomaly judgment result, and use the confirmed marking result as the arbitration fusion output. It should be noted that the arbitration fusion output is a unified set of anomaly judgment labels obtained after consistent arbitration of all residual current signal records with conflicting labels, which is used as the direct source of subsequent arbitration judgment results.

[0045] S153, output the arbitration fusion output as the arbitration judgment result; In a specific embodiment, the arbitration fusion output result obtained in step S152 is directly marked as the arbitration decision result and recorded in the local storage unit as the basis for subsequent model parameter backtracking updates. It should be noted that the arbitration ruling is the only valid ruling after resolving the conflict between the local and cloud-based anomaly rulings.

[0046] S154: When the model version number is inconsistent with the cloud model version number, the local model parameters are updated retrospectively based on the arbitration result.

[0047] In a specific embodiment, when the model version number is detected to be inconsistent with the cloud model version number, the arbitration judgment result is used as a supervision label and the corresponding residual current signal record are input to the backtracking optimization interface of the local model to perform incremental gradient update and complete the backtracking update of the local model parameters. It should be noted that the backtracking update of local model parameters enables the local model to synchronize the determination logic of the cloud model version number without completely relying on network stability, thereby maintaining the consistency of the determination results.

[0048] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0049] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0050] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0051] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0052] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0053] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A real-time remote protection method for residual current based on edge computing, characterized in that, The method includes: S11, based on the residual current signal data collected by the current sensors deployed on each power distribution branch and their corresponding collection timestamps, perform time sorting and differential calculation to generate a residual current signal record containing time interval values ​​and amplitude differences; S12, based on the residual current signal record and the model version number stored locally, extract the signal arrival delay value and time jitter value of each record, and fit the two types of data to generate time drift compensation parameters; S13, perform version consistency comparison based on time drift compensation parameters and cloud model version number returned by cloud interface to generate version consistency confidence value; at the same time, generate local anomaly judgment result based on residual current signal record and local model inference. S14, based on the local anomaly judgment result and the version consistency confidence value, a dual judgment process is triggered, the cloud model inference is called to obtain the cloud anomaly judgment result, and the consistency comparison between the local anomaly judgment result and the cloud anomaly judgment result is performed to generate a conflict marker record. S15: Based on the conflict marker record and the corresponding residual current signal record, perform arbitration fusion operation to generate arbitration judgment result. When the model version number is inconsistent with the cloud model version number, call the arbitration judgment result to perform local model parameter backtracking update.

2. The real-time remote protection method for residual current based on edge computing according to claim 1, characterized in that, The steps for generating residual current signal records are as follows: S111 synchronously acquires residual current signal data and acquisition timestamps through current sensors deployed on each power distribution branch; S112, Perform time sorting on the residual current signal data based on the acquisition timestamp to obtain sorted signal data; S113, Perform differential calculation based on the difference between adjacent data points of the sorted signal data to obtain the amplitude difference and time interval value; S114 records the amplitude difference and time interval in a structured manner and outputs the residual current signal record.

3. The real-time remote protection method for residual current based on edge computing according to claim 2, characterized in that, The logic for calculating the amplitude difference and time interval is as follows: S113.1, Extract the amplitude and time values ​​of two adjacent data points in the sorted signal data; S113.2, Subtract the amplitude of the previous data point from the amplitude of the next data point to generate an amplitude difference; S113.3, subtract the time value of the previous data point from the time value of the next data point to generate the time interval value.

4. The real-time remote protection method for residual current based on edge computing according to claim 3, characterized in that, The steps for generating time drift compensation parameters are as follows: S121, extract the first and last timestamps of each record in the residual current signal record to calculate the signal arrival time and obtain the signal arrival delay value; S122, calculate the time jitter value based on the time interval difference between adjacent records in the residual current signal record; S123, the signal arrival delay value and time jitter value are fitted into the time drift model to obtain the time drift compensation parameters.

5. The real-time remote protection method for residual current based on edge computing according to claim 4, characterized in that, The logic for obtaining the signal arrival delay value is as follows: S121.1, Obtain the start and end timestamps of each record in the residual current signal record; S121.2, subtract the start timestamp from the end timestamp to obtain the signal arrival delay value.

6. The real-time remote protection method for residual current based on edge computing according to claim 5, characterized in that, The steps to generate version consistency confidence values ​​are as follows: S131, obtain the model version number from the cloud via the cloud interface and read the model version number stored locally; S132, based on the model version number and the cloud model version number, performs string comparison and verification; S133 maps the comparison results to consistency level values ​​and outputs version consistency confidence values.

7. The real-time remote protection method for residual current based on edge computing according to claim 6, characterized in that, The steps to generate local anomaly determination results are as follows: S134, The residual current signal is recorded and input into the local model for inference to obtain the anomaly score value; S135: Perform classification operation based on the anomaly score and the model's built-in judgment threshold, and mark the classification result as the local anomaly judgment result.

8. A real-time remote protection method for residual current based on edge computing according to claim 7, characterized in that, The steps to generate a conflict marker record are as follows: S141, based on the local anomaly judgment result and the version consistency confidence value, the cloud model inference process is triggered synchronously. S142, input the residual current signal record into the cloud model to obtain the cloud anomaly judgment result; S143 compares the local anomaly determination results with the cloud anomaly determination results one by one and records the inconsistencies, outputting conflict marker records.

9. A real-time remote protection method for residual current based on edge computing according to claim 8, characterized in that, The steps for performing local model parameter backtracking updates by invoking the arbitration decision result are as follows: S151, locate the corresponding residual current signal record based on the collision marker record; S152, input the conflict mark record and the residual current signal record into the arbitration fusion calculation module to obtain the arbitration fusion output result; S153, output the arbitration fusion output as the arbitration judgment result; S154: When the model version number is inconsistent with the cloud model version number, the local model parameters are updated retrospectively based on the arbitration result.

Images