Multi-dimensional collaborative low-current grounding line selection accuracy optimization method

Abstract

The invention discloses a multi-dimensional collaborative small current grounding line selection accuracy optimization method. The method comprises the following main steps: S1, multi-source multi-dimensional data acquisition and working condition feature extraction; s2, carrying out load adaptive zero-sequence current reverse connection correction; s3, starting control of arc suppression coil adjustment delay adaptation; s4, intelligent line selection adaptive to the fault development stage; according to the invention, by forming a full-link collaborative optimization scheme from data acquisition to line selection output, the problems of misjudgment of reverse connection of zero-sequence current, interference of full working conditions of an arc suppression coil and inaccurate line selection of complex faults in a small-current grounding system are effectively solved; the applicability of the method is remarkably enhanced by quantifying the load factor, the dynamic parameter of the arc suppression coil, the fault development stage identifier and the fault type identifier feature; and due to the plug and play characteristic of the regional collaborative parameter library and the fault self-healing function of abnormal feedback, the operation and maintenance complexity is reduced, and convenience is provided for large-scale popularization and application of the scheme.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of ground fault detection of a power distribution network of a power system, and particularly relates to a multi-dimensional cooperative small-current grounding line selection accuracy optimization method. BACKGROUND

[0002] In a power system, a 35kV and below power distribution network is a key link connecting a power transmission network and a user side, and is directly related to power supply reliability and power utilization quality of terminal users; in order to reduce the impact of a fault current on equipment when a single-phase ground fault occurs, a small-current grounding system is generally used in the industry, and the core feature of the system is that when a single-phase ground fault occurs, the fault current is only a system capacitance current or a weak current compensated by an arc suppression coil, so that the electrical characteristics of a fault line and a non-fault line are not significantly different, which brings challenges to fault line selection, and the existing small-current grounding line selection method still has certain deficiencies.

[0003] Firstly, the load characteristics of the 35kV and below power distribution network vary significantly with user types and time: the total active load rate of an industrial substation is usually maintained at a high level during the day, and the load of a residential substation is concentrated in the evening peak load; load fluctuation causes distortion of a line zero sequence current waveform: when the load is high, the harmonic component of the line current increases, which may cover or distort the phase characteristics of the zero sequence current; when the load is low, the amplitude of the zero sequence current is further reduced, and the phase stability is decreased.

[0004] The existing zero sequence current reverse connection detection technology mostly adopts a fixed phase deviation threshold or a static fault template matching idea: the former presets a fixed phase deviation allowable value, and does not consider phase deviation caused by load fluctuation; the latter relies on a fixed waveform template of historical faults, and when the real-time load is significantly different from the load when the template is collected, the similarity calculation result is significantly deviated, and a normal line is easily misjudged as a reverse connection line, or an actual reverse connection line is missed; such load-independent detection logic causes a high misjudgment rate and a high omission rate of the reverse connection detection in a load fluctuation scenario, and it is difficult to meet the operation and maintenance requirements.

[0005] Secondly, in order to suppress an arc in a single-phase ground fault of a small-current grounding system, an automatic compensation type arc suppression coil is widely configured in a substation in the industry; the core function of the coil is to dynamically detect a system capacitance current after a fault occurs, to compensate the current by adjusting an inductance value of the coil, and to control the fault current in a safe range; however, due to the hysteresis characteristics of a coil core and the response speed of an adjusting mechanism, there is an adjusting delay time from fault triggering to stable adjustment of the arc suppression coil, and the delay is usually related to the type of the coil.

[0006] The existing line selection starting control technology is mostly based on static compensation degree design: or directly uses zero sequence voltage starting; or uses zero sequence current summation starting threshold with fixed coefficient correction, without distinguishing the compensation degree difference between the initial stage and the stable stage of the fault. In the initial stage of the fault, the coil compensation degree has not reached the target value, and the actual zero sequence current summation value may be lower than the preset threshold, resulting in missed starting. In the stable stage, if the initial stage threshold is still used, the summation value may be far higher than the threshold due to overcompensation, causing false starting and affecting the line selection reliability.

[0007] In addition, the single-phase grounding fault type of the small current grounding system is complex, and in addition to metallic grounding, the proportion of complex faults such as high resistance grounding and intermittent grounding is increasing. The electrical characteristics of such faults change throughout the life cycle: in the initial stage of the fault, the transient component dominates, and the waveform is strong. In the stable stage, the transient component decays, and the steady-state component becomes the main feature. If the fault is not handled in time, in the decay period, the transient component is basically eliminated, the steady-state component amplitude gradually decreases, and the waveform tends to be flat.

[0008] The existing intelligent line selection technology mostly uses the design idea of fixed feature dimension of fault type: for metallic grounding, it focuses on extracting high-frequency transient features; for high resistance grounding, it focuses on extracting steady-state features. However, in actual faults, the fault type and development stage are not completely separated, such as intermittent grounding, which repeatedly switches between transient and steady state, and the transient and stable stages of high resistance grounding are not clearly defined. The existing technology does not establish a dynamic adaptation relationship between the fault stage and the feature dimension, resulting in insufficient feature extraction in a certain stage due to fixed feature dimension, or feature fusion deviation due to static weight distribution, ultimately making it difficult to improve the line selection accuracy in complex fault scenarios.

[0009] Therefore, it is necessary to design a multi-dimensional collaborative small current grounding line selection accuracy optimization method. SUMMARY

[0010] The purpose of the present application is to provide a multi-dimensional collaborative small current grounding line selection accuracy optimization method to solve the problems of missed detection and misjudgment of zero sequence current detection caused by load fluctuation, delayed arc suppression coil adjustment causing missed and false starting of line selection starting control, and insufficient adaptation of feature dimension and weight in the dynamic change of electrical characteristics throughout the life cycle of complex faults, thereby improving the line selection accuracy and reliability of 35kV and below small current grounding systems in different scenarios such as single station independent operation and regional multi-station cluster operation, enhancing the adaptation ability of the method to different load characteristics, arc suppression coil working conditions and fault types, reducing the operation and maintenance complexity, and ensuring the stability of power supply of the distribution network.

[0011] To achieve the above purpose, the present application provides the following technical scheme: a multi-dimensional collaborative small current grounding line selection accuracy optimization method, comprising the following steps:

[0012] S1: Multi-source multi-dimensional data acquisition and working condition feature extraction: Collect fault recording data, real-time load data and equipment working condition data, extract load factors, arc suppression coil dynamic parameters and fault development stage characteristics from the collected data, and construct a correlation database;

[0013] S2: Zero sequence current reverse connection correction adaptive to load: Based on the real-time load data and fault recording data collected in S1, the fault similarity Calculate the dynamic adjustment of the phase deviation allowable value , implement symbol inversion and load compensation double correction on the line determined to be reversed, and generate corrected current data to eliminate phase deviation ;

[0014] S3: Start control adaptive to arc suppression coil regulation delay: Call the arc suppression coil regulation delay time extracted in S1 , calculate the compensation degree and introduce the delay factor in stages according to the fault stage characteristics , modify the start threshold value when the real-time zero sequence current sum reaches the modified threshold value, trigger the intelligent line selection process;

[0015] S4: Intelligent line selection adaptive to fault development stage: Receive the corrected current data output by S2 , switch the feature extraction dimension and weight distribution in stages in combination with the fault development stage characteristics of S1, input the multi-modal model for analysis, call the regional collaborative parameter library to output the final line selection result, and complete the accurate identification of the fault line;

[0016] Through the whole-link cooperation of multi-source data correlation, load adaptive correction, arc suppression coil delay adaptive start and fault stage line selection, the core line selection problem of the small current grounding system is solved; wherein, the correlation database of S1 provides a working condition adaptation basis for the subsequent steps, avoiding data isolation; the double correction of S2 solves the distortion of reverse connection detection caused by load fluctuation, the stage threshold modification of S3 avoids the start misjudgment / omission caused by arc suppression coil regulation delay, and the stage feature adaptation and regional collaboration of S4 improves the accuracy of line selection under complex faults, forming a closed loop optimization, significantly enhancing the adaptation ability of the line selection method to different loads, equipment working conditions and fault types, and ensuring the reliability and stability of the line selection result.

[0017] As a further technical solution of the application, in S1, the multi-source multi-dimensional data is specifically defined and the working condition feature extraction is:

[0018] Fault recording data: instantaneous value of zero sequence current of each line , zero sequence voltage of bus and sampling frequency , wherein , is the number of outgoing lines, is the sampling time;

[0019] Real-time load data: Collect the total active power data of the substation, and extract the load factor as the total active load rate wherein, , is the real-time total active power; is the rated total active power;

[0020] Device operating condition data: including arc suppression coil dynamic parameters: compensation degree and adjustment delay time wherein, is the real-time inductance current of the arc suppression coil, is the real-time capacitance current of the arc suppression coil, represents the time from fault triggering to stable adjustment of the coil;

[0021] Fault development stage characteristics: based on the transient energy change rate of the fault recording data, the fault development stage identifier is extracted :

[0022] In the initial stage, after the fault occurs ;

[0023] In the stable period, , is the stable duration;

[0024] In the decay period, .

[0025] As a further technical solution of the present application, in the S2, the specific logic of the load adaptive zero sequence current reverse correction is:

[0026] Load corrected fault similarity calculation: introduce load weight Correct the cosine similarity, the formula is:

[0027]

[0028] wherein, is the weight dynamically adjusted with the degree of load deviation from the rated value, the greater the load deviation, the smaller, is the matching window, is the instantaneous value of the zero sequence current reference waveform, is the similarity after load correction;

[0029] Dynamic phase deviation allowable value calculation:

[0030]

[0031] Wherein, is the basic deviation value, is the similarity correction coefficient, determined according to the system phase identification accuracy requirement, used to adjust the influence degree of the fault similarity on the phase deviation allowed value, is the load reference coefficient, determined based on the phase deviation characteristics under rated load, ensuring the rationality of the basic phase deviation, is the load influence coefficient, determined according to the sensitivity of system load rate change to phase deviation, reflecting the adjustment range of high and low load to phase deviation allowed value, is the final value, when the load is high , increases, adapting to waveform distortion under high load;

[0032] Symbol inversion and load compensation double correction:

[0033]

[0034] Wherein, is the corrected current, is the rated load rate reference value, is the load compensation coefficient, determined according to the system load current distortion characteristics, correcting the amplitude distortion of the reverse connection line under high load;

[0035] Through three-level optimization of load modified similarity, dynamic phase deviation and double correction, the interference problem of load fluctuation on zero sequence current reverse connection detection is effectively solved; the dynamic adjustment of load weight makes the similarity calculation adapt to different load scenarios, avoiding the deviation of fixed template matching; the dynamic phase deviation allowed value changes with the load rate , ensuring that the reverse connection line can still be accurately identified when the waveform is distorted under high load; symbol inversion and load compensation double correction not only corrects the phase deviation, but also compensates for the amplitude distortion under high load, avoiding the misjudgment or omission of reverse connection caused by current amplitude deviation, significantly improving the accuracy and adaptability of reverse connection detection under different load characteristics.

[0036] As a further technical solution of the present application, in the S3, the starting control logic of the delay adjustment adaptation is:

[0037] Stage compensation degree calculation:

[0038] In the initial stage of fault : real-time compensation degree is used to reflect the dynamic characteristics when the coil is not stable, and the formula is ;

[0039] stable period : adopt average compensation degree ;

[0040] Adjusting delay factor correction: define delay factor , fault duration less than , increase with time, greater than , keep constant, used to amplify the starting threshold of the initial stage of fault;

[0041] Dynamic starting threshold calculation:

[0042]

[0043] Wherein, is the phased compensation degree, is the basic safety factor, determined based on system reliability requirements, is the historical minimum summation value, is the compensation degree correction coefficient;

[0044] Starting decision: calculate real-time summation value , if , start line selection;

[0045] For the characteristics of arc suppression coil adjusting delay, through the coordinated design of phased compensation degree, delay factor and dynamic threshold, the problems of missing start at the initial stage of fault and false start at the stable period are solved; the phased compensation degree and adapt to the unstable and stable states of the coil respectively, avoiding the deviation caused by using a single compensation degree to calculate the threshold; the delay factor increases with time at the initial stage of fault, amplifies the starting threshold, ensures that the real-time summation value can still reach the threshold when the coil is unstable, avoiding missing start; the dynamic threshold combines the degree of compensation degree deviating from full compensation with the delay factor, which not only ensures the sensitivity of starting, but also avoids false start due to excessive compensation at the stable period, significantly improving the reliability of line selection and starting of automatic arc suppression coil substation.

[0046] As a further technical solution of the application, in the S4, the specific logic of fault stage adaptation line selection is:

[0047] Phased feature dimension switching:

[0048] Initial stage: extract high-frequency transient features, adapt to the characteristics of high-frequency components prominent at the initial stage of fault;

[0049] Stable period: extract low-frequency transient and steady-state features, adapt the characteristics of stable features;

[0050] Decay period: extract steady-state and low-frequency decay features, adapt the characteristics of transient component decay;

[0051] Characteristic frequency range formula: , wherein, is the rated frequency of the system, is the frequency adjustment coefficient;

[0052] Phased weight distribution:

[0053]

[0054]

[0055] , wherein, is the transient weight, is the initial transient basic weight, is the weight decay coefficient, which adapts to the law that the proportion of transient features decreases during fault development;

[0056] Regional coordination parameter calling: from the working condition and parameter sharing library of the regional edge node, the optimal threshold value of the same load rate and fault stage combination is called to avoid single station parameter deviation;

[0057] Through phased feature extraction, dynamic weight distribution and regional coordination, the problem of insufficient feature adaptation in the whole life cycle of complex faults is solved; phased feature dimension switching extracts effective features according to the electrical feature differences of fault stages , avoiding information omission caused by fixed feature dimension; the transient weight decreases with the fault stage, adapting to the natural change of the proportion of transient features, ensuring the rationality of feature fusion; regional coordination parameter calling avoids single station parameter debugging deviation by sharing the optimal threshold value of the same combination , especially adapting to complex fault scenarios such as intermittent grounding and high resistance grounding, significantly improving the accuracy and regional adaptability of line selection under different fault development stages.

[0058] As a further technical solution of the present application, in the S4, a stage correction coefficient is introduced in the calculation of the comprehensive feature value of the multi-modal model, which takes the maximum value in the stable stage, adapting to the characteristics that the features in this stage are the most stable, and the formula is:

[0059]

[0060] ​Wherein, is a transient feature matching value, is a steady-state feature matching value;

[0061] Preliminary route selection result determination: statistics of the number of windows satisfy The line is a candidate, is the total number of transient segment data windows, is a voting proportion threshold, combined with the same working condition route selection success rate of the region, the highest success rate is taken as the final preliminary result.

[0062] As a further technical solution of the application, it further includes S5: three-dimensional verification: data, working condition and region cooperation, specifically:

[0063] Data verification: calculate the phase difference between the route selection result line and the non-fault line , if and the amplitude is the largest, pass;

[0064] Working condition verification: at high load, the deviation between the corrected current and the original current should be within the preset allowable range, and at the initial stage of the arc extinguishing coil, the ratio of the real-time sum value to the starting threshold should not be less than the safety redundancy coefficient, pass;

[0065] Region verification: call the same working condition route selection record in the region library, if the historical success rate of this line under the same working condition is not lower than the average level of the region, pass.

[0066] As a further technical solution of the application, it further includes S6: parameter iterative optimization of multi-station cooperation, specifically:

[0067] Edge node summary: the edge computing node of the region summarizes the load , fault and route selection result data of each station every hour, and updates the regional shared library;

[0068] Hierarchical optimization:

[0069] Single station layer: after accumulating a preset number of fault data, fine-tune the station so that the reverse detection accuracy is not lower than the preset target;

[0070] Regional layer: periodically statistics the same working condition data of the region, optimize the feature threshold, so that the average route selection accuracy of the region is not lower than the preset target;

[0071] Abnormal feedback: if the route selection of a station fails more than a preset threshold number of times, the edge node automatically pushes the optimal parameters of similar working conditions in the same region to realize fault self-healing parameter adjustment.

[0072] As a further technical solution of the application, in the S6, the regional sharing library adopts a blockchain and edge storage architecture, each station data upload generates an unalterable timestamp, and the parameter optimization adopts federated learning to avoid original data leakage and adapt to the power system data security requirements.

[0073] As a further technical solution of the application, in the S1, the identification of the fault development stage adopts a transient energy change rate determination: calculating the transient energy , and dividing the stage according to the interval range of the energy change rate , the initial change rate is the largest, and the decay period is the smallest, so as to ensure that the stage identification accuracy is not less than a preset threshold.

[0074] Compared with the prior art, the beneficial effects of the multi-dimensional collaborative small current grounding line selection accuracy optimization method are:

[0075] Through forming a full-link collaborative optimization scheme from data acquisition to line selection output, the problems of misjudgment of zero sequence current connection, interference of arc suppression coil under all working conditions, and inaccurate line selection under complex faults in the small current grounding system are effectively solved, wherein the load adaptive zero sequence current connection correction corrects the fault similarity weight by the real-time load rate , dynamically adjusts the phase deviation allowable value , and introduces symbol inversion and load compensation double correction to avoid correction distortion caused by load fluctuation; the start control of the arc suppression coil adjustment delay adaptation calculates the compensation degree in stages, dynamically corrects the start threshold combined with the adjustment delay factor , solves the start misjudgment when the coil is not stable at the initial stage of the fault; the intelligent line selection of the fault development stage adaptation switches the feature extraction frequency in stages through fault stage identification , dynamically allocates transient feature weight and steady-state feature weight , adapts to the feature change in the whole life cycle of the fault, and improves the reliability and stability of the line selection result as a whole;

[0076] By quantifying the load factor, the dynamic parameters of the arc suppression coil, the fault development stage identification and the fault type identification features, different load characteristics, equipment working conditions and fault types can be adapted, without the need for a large number of customized adjustments for specific scenarios, significantly enhancing the applicability of the method;

[0077] Through multi-station collaborative parameter iterative optimization and regional sharing architecture based on blockchain and edge storage, the data security is ensured while the line selection continuous optimization capability is improved; the single station layer fine tunes the basic phase deviation allowable value Periodic optimization of regional layer threshold The hierarchical optimization mechanism enables the parameters to be continuously adapted to the actual working condition as historical data accumulates; the application of federated learning and blockchain technology avoids the risk of leakage of original recording wave data, and meets the data security requirements of the power system; in addition, the plug-and-play feature of the regional collaborative parameter library and the fault self-healing function of abnormal feedback reduce the operation and maintenance complexity, and facilitate the large-scale promotion and application of the scheme. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 is a method flowchart of the present application;

[0079] Figure 2 is a multi-modal intelligent route selection software module architecture and data processing flowchart;

[0080] Figure 3 is a comparison chart of fault line and non-fault line recording waveforms in a zero sequence current connection reverse state;

[0081] Figure 4 is a BMP waveform comparison chart after different window point division of zero sequence current recording data of a small current grounding fault. DETAILED DESCRIPTION

[0082] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0083] Please refer to the accompanying Figure 1 -attached Figure 4 An embodiment 1 provided by the present application is a multi-dimensional collaborative small current grounding route selection accuracy optimization method, comprising the following steps:

[0084] S1: Multi-source multi-dimensional data acquisition and working condition feature extraction: acquire fault recording data, real-time load data and equipment working condition data, extract load factors, arc suppression coil dynamic parameters and fault development stage features in the acquired data, and construct a correlation database;

[0085] The multi-source multi-dimensional data specific definition and working condition feature extraction are as follows:

[0086] Fault recording data: instantaneous value of zero sequence current of each line , zero sequence voltage of bus and sampling frequency , wherein, , is the number of outgoing lines, Tsample

[0087] Real-time load data: Collecting total active power data of substation, extracting load factor as total active load rate , , Ttotal Ttotal_set

[0088] Device operating condition data: Including arc suppression coil dynamic parameters: compensation degree and adjustment delay time , Lcoil Ccoil Tcoil

[0089] Fault development stage characteristics: Based on transient energy change rate of fault recording data, extracting fault development stage identification :

[0090] In the initial stage, after the fault occurs ;

[0091] In the stable period, , Tstable

[0092] In the decay period, ;

[0093] Identification of fault development stage Adopt transient energy change rate judgment: Calculate transient energy According to the interval range of energy change rate Divide the stage, the initial stage has the maximum change rate, and the decay period has the minimum change rate, to ensure that the stage identification accuracy is not less than the preset threshold;

[0094] S2: Load adaptive zero sequence current connection correction: Based on real-time load data and fault recording data collected by S1, through fault similarity Calculate dynamic adjustment of phase deviation allowable value , The line judged as connection is implemented symbol inversion and load compensation double correction, to generate corrected current data to eliminate phase deviation ;

[0095] The specific logic of load adaptive zero sequence current connection correction is:

[0096] Load corrected fault similarity calculation: Introducing load weight The modified cosine similarity is as follows:

[0097]

[0098] wherein, is a weight dynamically adjusted according to the degree of load deviation from the rated value, the greater the load deviation, the smaller, is a matching window, is an instantaneous value of the zero-sequence current reference waveform, is a load-modified similarity;

[0099] Dynamic phase deviation allowable value calculation:

[0100]

[0101] wherein, is a basic deviation value, is a similarity correction coefficient, determined according to the system phase identification accuracy requirement, used to adjust the influence degree of fault similarity on the phase deviation allowable value, is a load reference coefficient, determined based on the phase deviation characteristics under the rated load, to ensure the rationality of the basic phase deviation, is a load influence coefficient, determined according to the sensitivity of system load rate change to the phase deviation, reflecting the adjustment range of high and low loads on the phase deviation allowable value, is a final value, when the load is high, is large, increases, adapting to waveform distortion under high load;

[0102] Symbol inversion and load compensation double correction:

[0103]

[0104] wherein, is the corrected current, is a rated load rate reference value, is a load compensation coefficient, determined according to the system load current distortion characteristics, to correct the amplitude distortion of the reverse connection line under high load;

[0105] Through three-level optimization of load-modified similarity, dynamic phase deviation and double correction, the interference problem of load fluctuation on the zero-sequence current reverse connection detection is effectively solved; the dynamic adjustment of load weight makes the similarity calculation adapt to different load scenarios, avoiding the deviation of fixed template matching; the dynamic phase deviation allowable value changes with the load rate , ensuring that the reverse connection line can still be accurately identified when the waveform is distorted under high load; the symbol inversion and load compensation double correction not only corrects the phase deviation, but also Compensate the amplitude distortion under high load, avoid the misjudgment or missed judgment caused by the deviation of current amplitude, and significantly improve the accuracy and adaptability of the reverse connection detection under different load characteristics;

[0106] S3: Arc suppression coil adjustment delay adaptive start control: call the arc suppression coil adjustment delay time extracted in S1 With fault stage characteristics, calculate compensation degree and introduce delay factor in stages Correct the starting threshold value when the real-time zero sequence current sum value When the corrected threshold value is reached, trigger the intelligent line selection process.

[0107] The specific logic of the adjustment delay adaptive start control is:

[0108] Stage compensation degree calculation:

[0109] Fault initial stage : Use real-time compensation degree Reflect the dynamic characteristics when the coil is not stable, the formula is ;

[0110] Stable period : Use average compensation degree ;

[0111] Adjustment delay factor correction: define delay factor , when the fault duration is less than , it increases with time, and when it is greater than , it remains constant, used to amplify the starting threshold value in the initial stage of the fault.

[0112] Dynamic starting threshold value calculation:

[0113]

[0114] Where, is the stage compensation degree, is the basic safety factor, determined based on system reliability requirements, is the historical minimum sum value, is the compensation degree correction coefficient.

[0115] Start determination: calculate the real-time sum value , if , start line selection;

[0116] For the characteristics of arc suppression coil adjustment delay, through the coordinated design of stage compensation degree, delay factor and dynamic threshold, the problems of start missed judgment in the initial stage of fault and false start in the stable period are solved; stage compensation degree and adapt to the unstable and stable states of the coil respectively, avoiding the deviation caused by using a single compensation degree to calculate the threshold value; delay factor Amplification start threshold value increases over time in the initial stage of failure, ensuring that the sum value is real-time when the coil is not stable Still able to reach the threshold value, avoiding missed start; Dynamic threshold Combined with the degree of compensation deviation from full compensation With the delay factor, both the sensitivity of the start and the false start caused by excessive compensation in the stable period are avoided, significantly improving the reliability of the automatic arc suppression coil substation line selection start;

[0117] S4: Intelligent line selection adapted to the fault development stage: receive the corrected current data output by S2 , combined with the fault development stage characteristics of S1, the feature extraction dimension and weight distribution are switched in stages, and after inputting the multi-modal model for analysis, the regional collaborative parameter library is called to output the final line selection result, completing the accurate identification of the fault line;

[0118] The specific logic of the fault stage adapted line selection is:

[0119] Switching of feature dimensions in stages:

[0120] Initial stage: extract high-frequency transient features, adapt to the characteristics of high-frequency components prominent in the initial stage of failure;

[0121] Stable period: extract low-frequency transient and steady-state features, adapt to the characteristics of stable features;

[0122] Decay period: extract steady-state and low-frequency decay features, adapt to the characteristics of transient component decay;

[0123] Characteristic frequency range formula: , where, is the rated frequency of the system, is the frequency adjustment coefficient;

[0124] Stage weight distribution:

[0125]

[0126]

[0127] where, is the transient weight, is the initial transient base weight, is the weight decay coefficient, which adapts to the law of decreasing proportion of transient features in the development of failure;

[0128] Regional collaborative parameter call: from the working condition and parameter sharing library of the regional edge node, the same load rate and fault stage Optimal feature threshold of combination To avoid single-station parameter deviation;

[0129] By employing phased feature extraction, dynamic weight allocation, and regional collaboration, the problem of insufficient feature adaptation throughout the entire lifecycle of complex faults is addressed; the phased feature dimension switching is based on the fault stage. Based on the differences in electrical characteristics, effective features are extracted in a targeted manner to avoid information omissions caused by fixed feature dimensions; transient weights. As the fault stage decreases, the proportion of transient features naturally changes to ensure the rationality of feature fusion; regional collaborative parameter invocation is achieved through sharing. Optimal threshold of combination This avoids deviations in single-station parameter debugging, and is especially suitable for complex fault scenarios such as intermittent grounding and high-resistance grounding, significantly improving the accuracy and regional adaptability of line selection under different fault development stages;

[0130] The calculation of the comprehensive eigenvalues ​​of multimodal models introduces stage correction coefficients. The value is maximized when the condition is stable, reflecting the most stable characteristic of that stage. The formula is:

[0131]

[0132] in, For transient feature matching values, The steady-state feature matching value;

[0133] Preliminary route selection result determination: statistics Number of windows ,satisfy The route is a candidate. This represents the total number of transient segment data windows. The voting ratio threshold is used as a reference, and then the route with the highest success rate under the same working conditions in the region is ranked. The route with the highest success rate is taken as the final preliminary result.

[0134] By integrating multi-source data, adaptive load correction, arc suppression coil delay adaptation startup, and fault-stage line selection across the entire link, a targeted solution to the core line selection problem in low-current grounding systems is achieved. Specifically, the associated database in S1 provides a basis for operating condition adaptation in subsequent steps, avoiding data isolation; the dual correction in S2 solves the distortion of reverse connection detection caused by load fluctuations; the phased threshold correction in S3 avoids startup misjudgment / missed judgment caused by arc suppression coil adjustment delay; and the phased feature adaptation and regional collaboration in S4 improve the accuracy of line selection under complex faults. The whole system forms a closed-loop optimization, significantly enhancing the adaptability of the line selection method to different loads, equipment operating conditions, and fault types, and ensuring the reliability and stability of the line selection results.

[0135] Also includes S5: Three-dimensional verification: data, working condition and region coordination, specifically:

[0136] Data verification: Calculate the selected line result line and the phase difference of the non-fault line , if and the amplitude is the largest, pass;

[0137] Working condition verification: When the load is high, the deviation between the corrected current and the original current should be within the preset allowable range, and when the arc suppression coil is in the initial stage, the ratio of the real-time sum value to the starting threshold should not be less than the safety redundancy coefficient, pass;

[0138] Region verification: Call the same working condition line selection record in the region library, if the historical success rate of this line under the same working condition is not lower than the regional average level, pass;

[0139] Also includes S6: Parameter iterative optimization of multi-station coordination, specifically:

[0140] Edge node summary: The regional edge computing node summarizes the load , fault and line selection result data of each station every hour, and updates the regional shared library;

[0141] Hierarchical optimization:

[0142] Single station layer: After accumulating a preset number of fault data, fine-tune the station so that the reverse detection accuracy is not lower than the preset target;

[0143] Regional layer: Regularly statistics regional data under the same working condition, optimize the feature threshold, so that the average line selection accuracy of the region is not lower than the preset target;

[0144] Abnormal feedback: If the line selection of a station fails more than a preset threshold number of times, the edge node automatically pushes the optimal parameters of similar working conditions in the same region to realize fault self-healing parameter adjustment;

[0145] The regional shared library adopts a blockchain and edge storage architecture, and after uploading the data of each station, an unalterable timestamp is generated. Parameter optimization uses federated learning to avoid original data leakage and adapt to the data security requirements of the power system.

[0146] ​One embodiment provided by the application: the application object is a 35kV suburban substation, which undertakes mixed power supply task of 10kV industrial outgoing line 6 and 10kV residential outgoing line 6 of surrounding industrial park and residential community, 10kV side adopts neutral point grounding through automatic arc suppression coil small current grounding system, XHDC-10 type automatic arc suppression coil is configured, the adjustment response time is 0.12-0.18s, ZK300 type small current grounding line selection device, the historical line selection pain point is: high misjudgment rate of reverse detection in early morning peak, missed judgment in the initial stage of arc suppression coil adjustment, low accuracy of intermittent grounding line selection;

[0147] During the industrial early morning peak, the 10kV I bus of the substation has C-phase high resistance grounding fault, the on-site investigation confirms that the grounding resistance is about 150Ω, the fault is extinguished after 42s, which is the intermittent grounding characteristic, the line selection is completed by the method of the application, and the specific implementation process is as follows: One,

[0149] 1. Fault recording data acquisition: obtain COMTRADE format recording data through the FTP interface of the substation, the sampling frequency , the initial value of the zero sequence current of the 10kV I bus 12 outgoing lines is 0.012-0.038A, and the initial value of the zero sequence voltage of the bus is 28.5V;

[0150] 2. Real-time load data extraction: the total active power of the substation is displayed by the dispatching automation system , the rated total active power , and the load rate is calculated;

[0151] 3. Arc suppression coil dynamic parameter acquisition: the arc suppression coil controller outputs real-time inductance current , system capacitance current , compensation degree , and overcompensation state; the coil adjustment delay is confirmed through the manufacturer's technical manual ;

[0152] 4. Fault development stage identification: the transient energy is calculated , in the initial stage of the fault, the energy change rate is , when 0≤t≤0.15s, it is determined that ; when 0.15s<t≤0.15+0.3s=0.45s, the change rate drops to , it is determined that ; when t>0.45s, the change rate is , it is determined that ;

[0153] 5. Association database construction: the above recording data and load rate ​​Arc suppression coil parameters and fault stage Bind storage to local industrial computer by timestamp; Two,

[0155] 1. Fault similarity calculation:

[0156] Select the reference waveform of zero sequence current of historical metallic grounding fault , Set load weight , High load deviates from rated value, Lower, into the formula:

[0157]

[0158] 2. Dynamic phase deviation allowable value calculation:

[0159] Set the basic deviation , Similarity correction coefficient , Load reference coefficient , Load influence coefficient , Into the formula:

[0160]

[0161] 3. Double correction implementation:

[0162] Model identification 10kVI mother 1215 industrial park line 6 zero sequence current phase is the same as the fault line, determine the reverse, set the rated load rate reference , Load compensation coefficient , Into the correction formula:

[0163]

[0164] After correction, the zero sequence current phase of line 6 is reversed, and the amplitude is compensated for distortion under high load; Three,

[0166] 1. Compensation degree calculation in stages:

[0167] Fault initial stage , t=0.1s: real-time compensation degree ;

[0168] Stable period , t=0.2s: stable duration , Average compensation degree ;

[0169] 2. Delay factor correction:

[0170] Fault duration t=0.1s , Delay factor ;

[0171] 3. Dynamic threshold and start determination:

[0172] Set the basic safety factor , the historical minimum sum value , the compensation correction coefficient , substitute the formula:

[0173]

[0174] Calculate the real-time sum value , because , trigger the intelligent routing process; Four,

[0176] 1. Stage feature extraction:

[0177] : According to the formula , , , extract 1-3kHz high-frequency transient characteristics;

[0178] Transient weight , steady-state weight ;

[0179] 2. Regional collaborative parameter calling:

[0180] In the regional edge node shared library, the same - Optimal feature threshold value ;

[0181] 3. Multi-modal model analysis and result determination:

[0182] The corrected current data is converted into BMP image input Qwen2.5 VL model, and the transient feature matching value of line 6 , steady-state feature matching value , stage correction coefficient is the initial feature stability, and the comprehensive feature value is:

[0183]

[0184] The number of in 13 transient windows , the proportion , ; Combined with the regional same working condition routing record, the historical success rate of line 6 is 96.5%, and the regional average is 94%, so line 6 is determined as the final fault line; Five,

[0186] 1. Data verification: Phase difference between line 6 and non-faulty line 3 ,and To the maximum, pass;

[0187] 2. Operating condition verification: High load When correcting deviation Deviation rate ≤ 10% (3.4%); Initial stage Safety redundancy approved;

[0188] 3. Area Verification: Line 6 in - Under normal operating conditions, the historical success rate is 96.5%, which is higher than the regional average of 94%, thus passing the test. six,

[0190] After the fault is resolved, the regional edge nodes aggregate the station's data to the blockchain shared repository:

[0191] Single-station layer: After accumulating 50 fault data points, fine-tune the basic phase deviation. ;

[0192] Regional layer: Weekly optimization - Feature threshold .

[0193] In summary, this invention effectively solves the problems of misjudgment of zero-sequence current reversal, interference from arc suppression coils under all operating conditions, and inaccurate fault location in low-current grounding systems by forming a full-link collaborative optimization scheme from data acquisition to line selection output. Specifically, the load-adaptive zero-sequence current reversal correction is achieved through real-time load factor... Correcting fault similarity weights Dynamically adjust the allowable value of phase deviation Furthermore, it introduces a dual correction mechanism of sign reversal and load compensation to avoid correction distortion caused by load fluctuations; the start-up control of the arc suppression coil adjustment delay adaptation calculates the compensation degree in stages, combined with the adjustment delay factor. Dynamically adjust the startup threshold This solves the problem of misjudgment during startup when the coil is not stable in the early stages of a fault; the intelligent line selection adapted for the fault development stage uses fault stage identification. Phased switching of feature extraction frequency Dynamically allocate transient feature weights With steady-state characteristic weights It adapts to the characteristic changes throughout the entire fault lifecycle, thus improving the overall reliability and stability of the route selection results;

[0194] By quantifying load factors, arc suppression coil dynamic parameters, and fault development stage indicators and fault type identifier The characteristics can be adapted to different load characteristics, equipment working conditions and fault types, without the need for a large number of customized adjustments for specific scenarios, and the applicability of the method is significantly enhanced;

[0195] Through the parameter iterative optimization of multi-station cooperation and the regional sharing architecture of blockchain and edge storage, the data security is guaranteed while the line selection continuous optimization capability is improved; the single station layer fine tunes the basic phase deviation allowable value after accumulating a preset number of fault data , the regional layer optimizes the feature threshold regularly , the hierarchical optimization mechanism enables the parameters to be continuously adapted to the actual working conditions with the accumulation of historical data; the application of federated learning and blockchain technology avoids the risk of leakage of original recording wave data, and meets the data security requirements of the power system; in addition, the plug-and-play characteristics of the regional cooperative parameter library and the fault self-healing function of abnormal feedback reduce the operation and maintenance complexity, and provide convenience for the large-scale popularization and application of the scheme.

[0196] It is apparent for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, but can be implemented in other concrete forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all aspects as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and all changes falling within the meaning and range of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be considered as limiting the claims involved.

Claims

1. A multi-dimensional collaborative method for optimizing the accuracy of low-current grounding line selection, characterized in that: Comprise the following steps: S1: multi-source multi-dimensional data acquisition and working condition feature extraction: collect fault recording data, real-time load data and equipment working condition data, extract load factor, arc suppression coil dynamic parameters and fault development stage characteristics in the collected data, and construct a correlation database; S2: load-adaptive zero sequence current anti-phase correction: based on the real-time load data collected by S1 and the fault recording data, through the fault similarity Computing the dynamically adjusted phase deviation allowance , the line determined to be reversed is implemented with sign inversion and load compensation double correction, and corrected current data eliminating phase deviation is generated ; S3: Start control of arc suppression coil adjustment delay adaptation: call the arc suppression coil adjustment delay time extracted in S1 With the fault stage characteristics, the compensation degree is calculated in stages and a delay factor is introduced The start threshold is corrected when the real-time zero sequence current sum value When the corrected threshold is reached, the intelligent line selection process is triggered; S4: Intelligent route selection adapted to fault development stage: receive corrected current data output by S2 , combined with the fault development stage characteristics of S1, the feature extraction dimension and weight allocation are switched by stage, the multi-modal model is analyzed, the regional collaborative parameter library is called to output the final route selection result, and the accurate identification of the fault line is completed.

2. The multi-dimensional synergistic small current ground selection accuracy optimization method according to claim 1, characterized in that: In the S1, the multi-source multi-dimensional data is specifically defined and the working condition feature extraction is: fault recording data: instantaneous value of zero sequence current of each line , zero sequence voltage of bus and sampling frequency wherein, , is the number of outgoing lines, is the sampling time; Real-time load data: Collect the total active power data of the substation, and extract the load factor as the total active load rate wherein, , is the real-time total active power; is the rated total active power; Device operating data: includes arc suppression coil dynamic parameters: compensation degree and adjustment delay time wherein, is the real-time inductance current of the arc suppression coil, is the real-time capacitance current of the arc suppression coil, represents the time from fault triggering to stable adjustment of the coil; Fault development stage feature: transient energy rate of change based on fault recording data, extract fault development stage identifier : At the initial stage, after the occurrence of a failure ; , stable phase, , is the stable duration; , decay period, .

3. The multi-dimensional synergistic small current ground selection accuracy optimization method according to claim 1, characterized in that: In the S2, the specific logic of the load-adaptive zero sequence current reverse connection correction is: Load corrected fault similarity calculation: Introduce load weight Corrected cosine similarity, formula is: wherein, is a dynamically adjusted weight depending on the degree of load deviation from the rated value, the greater the load deviation, the smaller, is a matching window, is the instantaneous value of the zero-sequence current reference waveform, is the similarity after load correction; Dynamic phase deviation allowable value calculation: Wherein, is the basic deviation value, is the similarity correction coefficient, determined according to the system phase identification accuracy requirement, used to adjust the influence degree of the fault similarity on the phase deviation allowed value, is the load reference coefficient, determined based on the phase deviation characteristics under rated load, to ensure the rationality of the basic phase deviation, is the load influence coefficient, determined according to the sensitivity of the system load rate change to the phase deviation, reflecting the adjustment range of high and low load to the phase deviation allowed value, is the final value, when the load is high is large, is increased, which adapts to the waveform distortion under high load; Symbol inversion and load compensation double correction: wherein, is the corrected current, is the rated load rate reference value, is the load compensation coefficient, determined according to the system load current distortion characteristics.

4. The multi-dimensional synergistic small current ground selection accuracy optimization method according to claim 1, characterized in that: In the S3, the specific logic of the starting control of the adjustment delay adaptation is: Stage compensation degree calculation: Fault initial stage : Adopt real-time compensation degree , reflect dynamic characteristics when the coil is not stable, the formula is ; stationary phase : using average compensation degree ; Adjustment of delay factor correction: define delay factor , the fault duration is less than increases over time, greater than remains constant, for amplifying the start threshold of the initial failure; Dynamic starting threshold calculation: wherein, is a phase compensation degree, is a basic safety factor, determined based on system reliability requirements, is a historical minimum summation value, is a compensation degree correction factor; Start decision: compute running sum value If then start select line.

5. The multi-dimensional synergistic small current ground selection accuracy optimization method according to claim 1, characterized in that: In the S4, the specific logic of the fault stage adaptation of line selection is: Stage feature dimension switching: Initial: Extract high-frequency transient features, adapt to the characteristics of high-frequency components prominent in the initial stage of failure; Stable period: Extract low-frequency transient and steady-state features, and adapt the characteristics of the features tend to be stable; Decay period: Extracts the steady state and low frequency decay characteristics, fitting the characteristics of the transient component decay; Characteristic frequency range formula: wherein, is the system rated frequency, is the frequency regulation coefficient; Stage weight distribution: wherein, is a transient weight, is an initial transient base weight, is a weight decay coefficient, adapting the law that the proportion of transient characteristics in the failure development decreases. Regional coordination parameter calling: from the regional edge node working condition and parameter sharing library, call the same load rate in the same region and fault phase Optimal feature threshold of combination .

6. The multi-dimensional synergic small current ground selection accuracy optimization method according to claim 1, characterized in that: In the S4, the comprehensive characteristic value calculation of the multi-modal model introduces a stage correction coefficient , the maximum value when stable, adapt to the characteristics of the most stable characteristics in this stage, the formula is: wherein, is a transient feature match value, is a steady state feature match value; Preliminary route selection result determination: statistics the number of windows , meet the line as a candidate, total number of transient section data windows, is the voting proportion threshold, combined with the same working condition route selection success rate sorting, the highest success rate of the line is the final preliminary result.

7. The multi-dimensional synergic small current ground selection accuracy optimization method according to claim 1, characterized in that: Also comprising S5: three-dimensional verification: data, working condition and region coordination, specifically: Data validation: compute selected line result line Phase difference from non-fault line If and amplitude maximum, pass​ Working condition verification: when the load is high, the deviation between the corrected current and the original current should be within the preset allowable range, and when the arc suppression coil is in the initial stage, the ratio of the real-time summation value to the starting threshold should not be less than the safety redundancy coefficient; pass; Region verification: call the same working condition line selection records in the region library, if the historical success rate of the line in the same working condition is not lower than the average level of the region, pass.

8. The multi-dimensional synergic small current ground selection accuracy optimization method according to claim 1, characterized in that: Also comprising S6: multi-station coordinated parameter iterative optimization, specifically: Edge node aggregation: regional edge computing nodes aggregate station load every hour , faults and route selection result data, update regional shared library; Hierarchical optimization: Single station layer: every accumulation of a preset number of fault data, fine-tune the station Make the detection accuracy no less than the preset target; Region layer: periodically statistics the same working condition data in the region, optimize the feature threshold, so that the average line selection accuracy of the region is not lower than the preset target; Abnormal feedback: if the line selection of a station fails more than a preset threshold number of times, the edge node automatically pushes the optimal parameters of similar working conditions in the same region to realize fault self-healing parameter adjustment.

9. The multi-dimensional synergic small current ground selection accuracy optimization method according to claim 8, characterized in that: In the S6, the region shared library adopts a blockchain and edge storage architecture, and each station data uploaded generates a tamper-proof timestamp, and the parameter optimization adopts federated learning.

10. The multi-dimensional synergistic small current ground selection accuracy optimization method according to claim 1, characterized in that: In the S1, the fault development stage is identified by using the transient energy change rate determination: calculating the transient energy , and dividing the stage according to the interval range of the energy change rate . The initial change rate is maximum, and the decay period is minimum, so that the stage identification accuracy is ensured to be not lower than a preset threshold.

Images