Electric power engineering safety fence fault diagnosis method and system

Abstract

The invention relates to the technical field of power transmission and distribution safety, in particular to an electric power engineering safety fence fault diagnosis method and system, and the method comprises the steps: collecting a break-over voltage and a disturbance direction to construct a time sequence model, recognizing disturbance factors, screening stacked sections, extracting boundary disturbance features, and recognizing a resistance burst response. And fault diagnosis areas are divided by combining disturbance intensity analysis. According to the method, through multi-time-sequence acquisition of the conduction voltage and the disturbance direction, establishment of evolution modeling of the disturbance trend term, identification of the disturbance factor group and capture of the micro-disturbance change of the fence node in different time periods, the early-stage trend of fault occurrence is effectively described, and the limitation of dependence on single-time measurement is broken through. On the basis, the recognition process of the disturbance stacking section is combined with periodic accumulation and amplitude superposition of disturbance density, and the coverage capacity of the potential high-incidence area of the fault is improved. Furthermore, boundary disturbance behavior characteristics are extracted from an active graph constructed through periodic signals, and boundary trend capture of perturbation distribution is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission and distribution safety, and particularly to a fault diagnosis method and system for a safety fence in a power engineering project. Background Art

[0002] The technical field of power transmission and distribution safety includes the safety guarantee, monitoring, and control of transmission lines, distribution facilities, and their affiliated protective structures in the power system, so as to ensure the stable transmission of electric energy during the transmission and distribution process, prevent equipment and personnel from being electrically injured, and prevent the adverse effects of environmental factors on the system operation.

[0003] Among them, the fault diagnosis method for the safety fence in a power engineering project refers to a technical method for monitoring the status and identifying faults of the safety fence set at the site of a power transmission and distribution project to isolate the electrical equipment and the personnel passage area. By aiming at the safety risks caused by external force damage, abnormal grounding, loose connection, or metal corrosion of the fence structure, methods such as resistance continuity detection, potential difference measurement, and fence node connectivity judgment are used for fault determination.

[0004] In the prior art, based on resistance continuity detection, potential difference measurement, and node connectivity judgment, there are defects in relying on static index values for fence status monitoring, and the continuous evolution process of electrical safety risks in the time dimension is not covered. Most of its operation logics are intermittent detections, with insufficient capture of the instantaneous states of disturbance changes, and it is difficult to achieve dynamic tracking of the development of disturbances. Especially in the early stage of minor damage, the voltage and resistance parameters may not have exceeded the preset threshold, resulting in the inability to accurately detect potential abnormalities. For example, in the case of periodic small disturbances to the fence, a single measurement may show a normal state, but after multiple disturbances accumulate, structural fatigue is caused, and such risks are difficult to be identified by traditional means. In addition, due to the lack of periodic statistics and trend modeling of resistance fluctuation behaviors in the existing methods, the weak abnormal linkages between nodes cannot form effective correlation analysis, and the hidden channels of cross-node fault propagation are easily ignored. In terms of spatial distribution, most diagnostic strategies are based on independent node judgments, and a comprehensive map structure of cross-node interactions is not constructed, resulting in fuzzy regional fault location and too large repair ranges, reducing the emergency repair efficiency and the accuracy of resource allocation. The above deficiencies are prone to misjudgment and missed judgment in high-load and complex layout scenarios, affecting the stability of system operation and the safety guarantee of personnel operations. Summary of the Invention

[0005] The purpose of the present invention is to solve the drawbacks existing in the prior art, and to propose a fault diagnosis method and system for a safety fence in a power engineering project.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A fault diagnosis method for a safety fence in a power engineering project includes the following steps. S1: Obtain the conduction voltages and disturbance directions at multiple time points of the fence nodes through the conduction of the fence, perform time-series evolution modeling of the disturbance trend terms, and identify the disturbance factor groups of the fence nodes. S2: According to the disturbance change conditions of the disturbance factors in the disturbance factor group of the fence nodes, judge the aggregation trend of the conduction voltage sequence and screen the corresponding disturbance stacking sections. S3: Obtain the periodic signal sequence of the disturbance stacking section, construct a critical domain activity graph, and identify the characteristics of the boundary disturbance behavior sections formed in the graph. S4: Obtain the resistance response value sequence of the fence nodes corresponding to the characteristics of the boundary disturbance behavior sections and perform boundary response behavior identification to capture the boundary sudden response set. S5: Obtain the fence nodes corresponding to the boundary sudden response set, jointly analyze the resistance behavior parameter group and the disturbance intensity level of the target fence nodes, and divide the fence fault diagnosis area.

[0007] The improvements of the present invention are as follows: the disturbance factor group of the fence nodes is specifically the disturbance amplitude change rate, the disturbance direction flip frequency, and the disturbance trend slope; the disturbance stacking sections include the disturbance density peak section, the disturbance amplitude superposition interval, and the disturbance period concentration area; the characteristics of the boundary disturbance behavior sections include the signal intensity extreme difference area, the direction switching concentration section, and the disturbance duration frame section; the boundary sudden response set is specifically the abnormal resistance jump fence nodes, the non-stable resistance rebound fence nodes, and the resistance jump continuous abnormal fence nodes; the fence fault diagnosis area includes the fault boundary fence node group, the diagnostic space position partition, and the joint feature concentrated distribution area.

[0008] The improvements of the present invention are as follows: the specific steps of obtaining the conduction voltages and disturbance directions at multiple time points of the fence nodes through the conduction of the fence, performing time-series evolution modeling of the disturbance trend terms, and identifying the disturbance factor groups of the fence nodes are as follows: S101: Obtain the measured conduction voltage values, current change rates, and signal direction vectors of each fence conduction fence node at the power engineering site during consecutive sampling periods. Based on the measured conduction voltage values, construct a conduction voltage sequence according to the sampling order. Based on the signal direction vectors, calculate the direction change angle sequence between adjacent periods, and generate a voltage and direction input data group. S102: Based on the voltage and direction input data group, calculate the periodic disturbance amplitude change rate for the conduction voltage sequence, judge the number of direction changes within a specified time for the direction change angle sequence, and then pair and recombine the two into a continuous sequence structure to generate a disturbance trend modeling input vector set. S103: Continuously process the data pairs of each fence node in the modeling input vector set according to the disturbance trend, perform time trend series modeling using the autoregressive moving average algorithm, extract the disturbance amplitude change rate and disturbance direction flip frequency of each group of fence nodes in each period, and obtain the fence node disturbance factor group.

[0009] The improvement of the present invention is that the specific steps for judging the aggregation trend of the conduction voltage sequence and screening the corresponding disturbance stacking section according to the disturbance change situation of the disturbance factors in the fence node disturbance factor group are as follows: S201: Call the disturbance amplitude change rate and disturbance direction flip frequency corresponding to each fence node in the fence node disturbance factor group, and combine the periodic arrangement structure of the same fence node in the conduction voltage sequence to construct a periodic signal sequence set; S202: According to the signal amplitude of each fence node in the periodic signal sequence set, perform per-period superposition on the continuous-period disturbance amplitude, calculate the disturbance amplitude accumulation of each fence node in each period, and arrange them in the sampling order to form a disturbance accumulation curve. Perform density statistics on the curve to calibrate the disturbance density peak interval; S203: Call the disturbance amplitude accumulation rate of the corresponding fence node in the disturbance density peak interval, and perform a ratio judgment with the average value of the disturbance amplitude change rate in the fence node disturbance factor group. If the former is twice or more than the latter, screen the corresponding time period and output it as the disturbance stacking section.

[0010] The improvement of the present invention is that the specific steps for obtaining the periodic signal sequence of the disturbance stacking section, constructing the critical domain activity graph, and identifying the boundary disturbance behavior section characteristics formed in the graph are as follows: S301: Call the periodic signal sequence corresponding to each fence node in the disturbance stacking section, extract the signal amplitude value and direction change angle in each period, pair them as equally spaced signal observation points in chronological order, and perform joint division of amplitude and direction based on the time period range to generate a synchronous statistical input segment set; S302: According to the signal amplitude and direction angle data of the observation points in the synchronous statistical input segment set, calculate the signal intensity range, direction switching frequency, and disturbance duration frame number of adjacent periods respectively, and normalize and combine the three structural values into a multi-dimensional statistical index to construct a stacking section structure parameter group; S303: Call the stacking section structure parameter group, construct a two-dimensional density distribution graph with the signal intensity range as the horizontal axis and the direction switching frequency as the vertical axis, identify the area with the largest statistical density in the graph as the disturbance focusing block, and output the boundary disturbance behavior section characteristics based on the distribution range of the focusing block.

[0011] The improvement of the present invention is as follows. The specific steps for obtaining the resistance response value sequence of the fence nodes corresponding to the boundary disturbance behavior section features and performing boundary response behavior recognition, and capturing the boundary sudden response set are as follows: S401: Call the fence node numbers corresponding to the boundary disturbance behavior section features, obtain the resistance response value sequence of the fence nodes within the corresponding period, extract the resistance change interval and the minimum response value in each period, and generate a fence node resistance sequence set; S402: According to the fence node resistance sequence set, calculate the resistance transient amplitude, resistance rebound duration, and jump duration period within the period for each fence node, respectively using the amplitude difference, fluctuation recovery duration, and abnormal period number as calculation items, merge multiple data vectors, and construct a fence node resistance behavior parameter group; S403: Call the feature vectors of each fence node in the fence node resistance behavior parameter group, judge the vector clustering distribution based on distance similarity, and identify the fence nodes deviating from the normal clustering center through the isolation forest algorithm to screen the boundary sudden response set.

[0012] The improvement of the present invention is as follows. The specific steps for obtaining the fence nodes corresponding to the boundary sudden response set, jointly analyzing the target fence node resistance behavior parameter group and the disturbance intensity level, and dividing the fence fault diagnosis area are as follows: S501: Call the fence nodes in the boundary sudden response set, obtain the fence node resistance behavior parameter group and the corresponding disturbance intensity level of the fence nodes in the previous period, and pair the two according to the fence node numbers to establish a fence node response index combination; S502: According to the resistance transient amplitude and disturbance intensity level of each fence node in the node response index combination, arrange them in a spatial grid structure based on the relative coordinate positions of the nodes in the fence layout, perform gradient cross calculation on the grid points and superimpose the mapping values to generate a fence space joint response map; S503: Call all the grid points in the fence space joint response map, identify the set of fence nodes whose disturbance intensity level exceeds one time the average value of all nodes in the area where the resistance rebound trend slows down, perform attribution determination on the set space range and output it as the fence fault diagnosis area.

[0013] A power engineering safety fence fault diagnosis system, the system includes: A disturbance factor identification module, which obtains the conduction voltage and disturbance direction of the fence at multiple time points through the fence-conducting fence nodes, performs time-series evolution modeling of the disturbance trend items, and identifies the fence node disturbance factor group; A disturbance aggregation screening module, which judges the aggregation trend of the conduction voltage sequence and screens the corresponding disturbance stacking section according to the disturbance change situation of the disturbance factors in the fence node disturbance factor group; Boundary behavior extraction module, which obtains the periodic signal sequence of the perturbation stacking section, constructs a critical domain activity graph, and identifies the boundary perturbation behavior section features formed in the graph; Resistance response identification module, which obtains the resistance response value sequence of the fence nodes corresponding to the boundary perturbation behavior section features and performs boundary response behavior identification to capture the boundary sudden response set; Fault diagnosis generation module, which obtains the fence nodes corresponding to the boundary sudden response set, jointly analyzes the resistance behavior parameter group of the target fence nodes and the perturbation intensity level, and divides the fence fault diagnosis area.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by collecting the multi-time series of the conduction voltage and the perturbation direction and establishing the evolution model of the perturbation trend term, the perturbation factor group is identified, the micro-perturbation changes of the fence nodes at different times are captured, the early trend of the fault occurrence is effectively depicted, and the limitation of only relying on single measurement is broken through. On this basis, the identification process of the perturbation stacking section combines the periodic accumulation of the perturbation density and the amplitude superposition to form the active extraction of the abnormal aggregation paragraph, improving the coverage ability of the potential high-incidence area of the fault. Further, the boundary perturbation behavior features are extracted from the activity graph constructed by the periodic signal, realizing the capture of the boundary trend of the micro-perturbation distribution, making the resistance behavior analysis no longer limited to the comparison of fixed indicators, but having the global trend judgment at the map level. When performing sudden behavior identification on the resistance response of the boundary nodes, a response feature vector is constructed by combining multi-dimensional parameters such as the jump amplitude, the rebound duration, and the continuous period, and then the isolation forest model is used to screen the abnormal points deviating from the clustering center, effectively avoiding the false alarms and missed alarms caused by unreasonable setting of the rule threshold. Finally, the spatial joint response map constructed by combining the resistance behavior and the perturbation level strengthens the analysis ability of the cross-influence between multiple nodes, realizes the three-dimensional division of the boundary buffer area in the gradient calculation, and accurately delimits the boundary of the diagnosis area. Brief Description of the Drawings

[0015] Figure 1 It is the method flow chart of the present invention; Figure 2 It is the schematic diagram of the refined process of step S1 of the present invention; Figure 3 It is the schematic diagram of the refined process of step S2 of the present invention; Figure 4 It is the schematic diagram of the refined process of step S3 of the present invention; Figure 5 It is the schematic diagram of the refined process of step S4 of the present invention; Figure 6 It is the schematic diagram of the refined process of step S5 of the present invention; Figure 7 It is the system module diagram of the present invention. Detailed implementation mode

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0018] Please refer to Figure 1 , the present invention provides a technical solution: a method for diagnosing faults in a safety fence for power engineering, including the following steps: S1: Obtain the conduction voltage and disturbance direction of the fence at multiple time points through the fence conduction fence nodes, perform time series evolution modeling of the disturbance trend term, and identify the fence node disturbance factor group; S2: According to the disturbance change conditions of the disturbance factors in the fence node disturbance factor group, judge the aggregation trend of the conduction voltage sequence and screen the corresponding disturbance stacking sections; S3: Obtain the periodic signal sequence of the disturbance stacking section, construct a critical domain activity graph and identify the characteristics of the boundary disturbance behavior section formed in the graph; S4: Obtain the resistance response value sequence of the fence node corresponding to the boundary disturbance behavior section characteristics and perform boundary response behavior identification to capture the boundary sudden response set; S5: Obtain the fence nodes corresponding to the boundary sudden response set, jointly analyze the resistance behavior parameter group and disturbance intensity level of the target fence node, and divide the fence fault diagnosis area; The fence node disturbance factor group specifically includes the disturbance amplitude change rate, the disturbance direction flip frequency, and the disturbance trend slope. The disturbance stacking section includes the disturbance density peak section, the disturbance amplitude superposition interval, and the disturbance period concentration area. The boundary disturbance behavior section characteristics include the signal strength range area, the direction switching concentration section, and the disturbance duration frame section. The boundary sudden response set specifically includes the fence nodes with abnormal resistance jumps, the fence nodes with non-stable resistance rebounds, and the fence nodes with continuous abnormal resistance jumps. The fence fault diagnosis area includes the fault boundary fence node group, the diagnostic space position partition, and the joint feature concentrated distribution area.

[0019] Please refer to Figure 2 , and the specific steps for obtaining the conduction voltage and disturbance direction at multiple time points of the fence through the fence conduction node, and performing the time series evolution modeling of the disturbance trend term, and identifying the specific disturbance factor group of the fence node are as follows: S101: Obtain the measured conduction voltage value, current change rate, and signal direction vector of each fence conduction node at the power engineering site within a continuous sampling period. Based on the measured conduction voltage value, construct a conduction voltage sequence according to the sampling order. Based on the signal direction vector, calculate the direction change angle sequence between adjacent periods to generate a voltage and direction input data group; When obtaining the measured conduction voltage value, current change rate, and signal direction vector of each fence conduction node at the power engineering site within a continuous sampling period, first, a voltage sampling module for collecting conduction voltage, a current sensor (such as a Hall element), and a three-dimensional direction sensing module need to be installed at each conduction fence node. The device continuously records voltage, current, and direction data by setting a sampling period (for example, collecting once every 2 seconds). Within each period, the voltage sampling module records the instantaneous voltage value such as 218V, and the current sensor outputs the current value such as 5.1A. The current change rate can be calculated for two consecutive periods using the formula: Current change rate ; where: : The current value measured in the previous period, for example, 5.1A; : The current value measured in the current period, for example, 4.6A; : The sampling period time interval (seconds), set to 2 seconds; , indicating that the current decreases at a rate of 0.25A per second within this period.

[0020] Next, the direction vector outputs spatial vector data by the three-dimensional direction sensing module in each period. This vector contains three components , , , respectively representing the components of the direction on the three coordinate axes (east-west, north-south, vertical direction) within this period. The direction vectors , represent the sensed direction vectors in two adjacent periods. The direction change angle between two periods can be calculated through the vector included angle formula : ; The parameter explanations are as follows: : The component of the direction vector on the X-axis in Period 1, indicating the degree of the direction towards the east within this period; : The component of the direction vector on the Y-axis in Period 1, indicating the degree of the direction towards the north within this period; : The component of the direction vector on the Z-axis in Period 1, indicating the degree of the direction upwards within this period; , , : They are the components of the direction vector in the X, Y, and Z axes in Period 2, corresponding to the above; dot product : A measure reflecting the consistency of the direction vectors between two periods; norm , : They respectively represent the lengths of two direction vectors and are used to standardize the results.

[0021] Let the direction vector of Period 1 be , and the direction vector of Period 2 be , indicating that both periods are in the plane direction.

[0022] Dot product calculation: ; Norm calculation: , ; Angle calculation: ; It shows that the direction between adjacent periods has changed by 8.1 degrees. If this angle is higher than the preset threshold, it is recorded as an effective change.

[0023] The voltage data is arranged in a sequence according to the sampling order. For example, the voltages of 10 consecutive periods are 218V, 219V, 220V, 221V, 220V, 218V, 217V, 219V, 220V, 222V, and the sequence of direction change angles is 8.1°, 12.3°, 5.0°, etc. The two form data pairs such as: (218, 0°), (219, 8.1°), (220, 12.3°), and sequentially constitute the voltage and direction input data set.

[0024] S102: Based on the voltage and direction input data set, calculate the change rate of the periodic disturbance amplitude for the conduction voltage sequence, judge the number of direction changes within the specified time for the direction change angle sequence, and then pair and recombine the two into a continuous sequence structure to generate the input vector set for disturbance trend modeling; Based on the voltage and direction input data set, in the process of calculating the change rate of the periodic disturbance amplitude for the conduction voltage sequence, it is necessary to first determine the disturbance amplitude, which is the absolute value of the difference between the voltage in the current period and the previous period, and then calculate the change rate of the disturbance amplitude as the difference between adjacent disturbance amplitudes divided by the time interval. For example, if the voltages of Period 1 and 2 are 218V and 220V respectively, the disturbance amplitude is , the voltages of Period 2 and 3 are 220V and 222V, and the disturbance amplitude is also 2V, and the change rate is . If the voltage of Period 3 and 4 becomes 224V, the disturbance is The change rate remains 0. If there are fluctuations, they are reflected in this rate value. The number of direction changes is judged by setting an angle threshold, such as 30°. If the direction angle changes by more than 30°, it is considered a direction change. For example, if the direction angle sequence is 15°, 35°, 5°, 40°, where 35° and 40° exceed the threshold, then the number of direction changes within this time period is 2. The sequence of the number of direction changes is the cumulative value of the direction changes in all consecutive time periods. Subsequently, the change rate of the periodic perturbation amplitude and the number of direction changes are aligned and paired in time to form a structured continuous sequence. For example, if the change rate of the first-period perturbation is 0.5 V / s and the number of direction changes is 1, then a corresponding vector is generated Finally, a complete set of input vectors for the perturbation trend modeling is formed.

[0025] S103: Perform continuous periodic input processing on the data pairs of each fence node in the input vector set for the perturbation trend modeling. Use the autoregressive moving average algorithm to perform time trend sequence modeling, extract the perturbation amplitude change rate and the perturbation direction flip frequency of each group of fence nodes in each period, and obtain the group of perturbation factors of the fence nodes; When performing continuous periodic input processing on the data pairs of each fence node in the input vector set for the perturbation trend modeling, use the autoregressive moving average algorithm to perform time trend modeling on the perturbation amplitude change rate and the direction flip frequency. First, it is necessary to define the input vector for the perturbation trend modeling as , where represents the perturbation amplitude change rate of the th sampling period, represents the direction flip frequency of the corresponding period. These data are respectively calculated from the measured voltage value and the direction angle sequence in Paragraph 2. The calculation of the perturbation amplitude change rate needs to use the on-state voltage values of the current period , the previous period and the period before the previous period . The defined formula is as follows: ; Among them: : The measured on-state voltage value of the th period, obtained through the voltage sampling module; : The on-state voltage value of the th period, representing the sampling result of the previous period; : The on-state voltage value of the th period, representing the result of the period before the previous period; : The duration of the sampling period, set to 2 seconds on-site, reflecting the change rate of the voltage perturbation on the time scale; : The serial number of the period currently being modeled and analyzed, such as the 10th period; : The period before the current period, that is, the 9th period; : The second cycle before the current cycle, i.e., the 8th cycle; For example: , , , substitute into the formula: , if the voltage in the next cycle is 226V, then: , indicating disturbance acceleration.

[0026] Direction reversal frequency is calculated as follows: , where: : The number of times the direction angle change exceeds the set threshold during the period time window (recommended to be ); : The statistical duration window, generally taken as 10 seconds; : In units of Hz, indicating the frequency of rapid direction changes; For example, if a total of 5 direction changes with an angle greater than 30° are detected in the cycle, then: , next, combine with to form the corresponding vector sequence , perform modeling prediction on this time series, and use the sliding window method for first-order autoregressive modeling (AR(1)), with the window width set to , based on the setting of the common short-term local fluctuation cycle length in the power disturbance scenario, the model structure is as follows: , where: : The constant offset term, reflecting the stable disturbance background, which can be obtained by fitting the data of the first 5 cycles. For example, set the initial ; : The autoregressive coefficient, reflecting the influence degree of the disturbance in the previous cycle on the current cycle, empirically set to 0.8; : The white noise error, following a normal distribution with a mean of 0 ; : The disturbance change rate in the previous cycle, from the vector sequence; : The predicted value of the current cycle.

[0027] For example, if the input is: , , , , , then through least squares fitting, can be obtained, and the prediction for the next cycle is: , the direction reversal frequency sequence Sliding average modeling can be performed in parallel. If exponential weighted moving average is adopted: , set parameters: : Empirical value, reflecting the proportion of current cycle data; , then: , finally, the change rate of the predicted disturbance amplitude and the direction reversal frequency are combined to form a disturbance factor group. For example, the prediction result of the 11th cycle is , which is used to dynamically describe the disturbance trend characteristics of the power fence node.

[0028] Please refer to Figure 3 , according to the disturbance change conditions of the disturbance factors in the disturbance factor group of the fence node, the specific steps to judge the aggregation trend of the conduction voltage sequence and screen the corresponding disturbance stacking section are as follows: S201: Call the change rate of the disturbance amplitude and the disturbance direction reversal frequency corresponding to each fence node in the disturbance factor group of the fence node, and combine the periodic arrangement structure of the same fence node in the conduction voltage sequence to construct a periodic signal sequence set; The rate of change of the disturbance amplitude corresponding to the fence node and the frequency of the disturbance direction flip are extracted in the previous stage of modeling. Essentially, they reflect the fluctuation intensity of the electrical disturbance and the activity level of the direction change of each node over a period of time. The rate of change of the disturbance amplitude is calculated from the difference in the amplitudes of the continuous periodic voltage changes. The specific method is as follows: in two adjacent periods, first calculate the absolute difference of the voltage values as the "disturbance amplitude", then compare this disturbance amplitude with the disturbance amplitude of the previous period, and divide the difference by the time of the sampling period to obtain the speed of the disturbance change. The magnitude of this value can be directly obtained from the voltage values collected by the sensor. The device is set to sample once every 2 seconds, and all changes are arranged in chronological order, so the calculation is continuous. The frequency of the disturbance direction flip is obtained by the direction sensing module recording the angle of the change of the direction vector in each period to identify whether there is a significant deviation in the direction. The specific method is to use an angle threshold, such as 30 degrees. As long as the angle between the direction vectors of two consecutive periods is greater than this threshold, it is considered that a direction flip has occurred. Count the number of such flips within a fixed time window and then divide by the window duration to obtain the flip frequency. For example, if there are 3 flips in 10 seconds, the frequency is 0.3 Hz. This frequency can indicate whether the fence node has experienced a significant disturbance in the direction during this period. The periodic signal sequence is a composite data arranged in chronological order. Each record contains three types of data of a certain node in a certain period: the voltage of this period, the rate of change of the disturbance amplitude in this period, and the frequency of the direction flip in this period. For a simple example, if the voltage of a certain node in the first period is 220V, the rate of change is 0.5, and the flip frequency is 0.2, these three pieces of data will be added to the sequence as a data unit; the three pieces of data of the next period are connected in sequence until all the periodic data of this node are arranged. This construction process ensures that there is evidence for subsequent analysis of the disturbance trend and also ensures the synchronization and alignment between different types of signals.

[0029] S202: According to the signal amplitudes of each fence node in the periodic signal sequence set, perform period-by-period superposition on the continuous periodic disturbance amplitudes, calculate the cumulative disturbance amplitude of each fence node in each period, arrange them in the sampling order to form a disturbance cumulative curve, perform density statistics on the curve, and calibrate the disturbance density peak interval; After the periodic signal sequence is constructed, the disturbance amplitude is cumulatively calculated periodically to reveal the development trend and change density of the disturbance. The cumulative calculation of the disturbance amplitude is a step-by-step accumulation process. Starting from the first period, the change rate of the disturbance amplitude within each period is added up. For example, if the change rate in the first period is 0.5 and in the second period is 0.7, then the cumulative value in the second period is 0.5 + 0.7 = 1.2, and so on. This approach essentially simulates the process of disturbance accumulation over time. In power engineering, local disturbances often occur due to frequent changes such as equipment startup and shutdown, and load fluctuations. If there are obvious disturbances at a certain node for several consecutive periods, the cumulative value will increase rapidly, thus showing the characteristic of "cumulative steep rise". Arranging the cumulative values in the order of the period numbers forms the disturbance cumulative curve. The shape of this curve can reveal which time periods the disturbances gather and which time periods are relatively stable. To quantitatively judge these gathering regions, the system will conduct density statistics on the curve. The method is to divide the time axis into multiple equal-length segments (such as each segment being 10 seconds long), and then count the number of cumulative increases in the disturbance within each segment, that is, the "disturbance growth density" within that segment. If the cumulative increase in a segment is relatively large, such as the growth amplitude exceeding 2.0, while in other segments it is only 0.8, 1.2, etc., obviously this segment is a high-density area. During the statistical process, by comparing the total growth amounts in each segment, the time periods with prominent growth amplitudes are extracted and marked as the disturbance density peak intervals.

[0030] S203: Call the disturbance amplitude cumulative rate of the corresponding fence node in the disturbance density peak interval, and perform a ratio judgment with the average value of the disturbance amplitude change rates in the disturbance factor group of the fence node. If the former is twice or more than the latter, then screen the corresponding time period and output it as the disturbance stacking section; After the density peak intervals are calibrated, the system will further determine whether these regions are significant enough to confirm whether they constitute a "perturbation stacking section". The specific method is to call the cumulative rate of perturbation amplitude within these density peak intervals, that is, divide the total cumulative perturbation within a certain interval by the duration of this interval to obtain the average growth rate, and then compare it with the average rate of change of perturbation amplitude in the historical cycle of this node. This "comparison" is not a simple difference, but a multiple judgment. If the cumulative rate within the density interval is twice or more than the historical average, it is determined as "abnormal stacking", because this indicates that the perturbation growth rate is much faster than the normal state in a short period of time. For example: The average rate of change of perturbation amplitude of a certain node in the entire monitoring cycle is 0.3. A certain interval lasts for 5 seconds, and the total cumulative perturbation is 3.2. Then the perturbation rate of this section is 3.2 divided by 5, which is equal to 0.64. Comparing 0.64 with 0.3, it is found to be 2.13 times, exceeding the "two times" standard. Therefore, this time period will be output as a perturbation stacking section. If the rate of a certain section is only 0.5, which is 1.66 times compared with the average value of 0.3, it is less than two times and will not be screened. This judgment method has clear boundary conditions and is also convenient for the subsequent system to extract high-risk time periods according to fixed rules for manual inspection or system automatic response processing.

[0031] Please refer to Figure 4 , to obtain the periodic signal sequence of the perturbation stacking section, the specific steps for constructing the critical domain active graph and identifying the characteristics of the boundary perturbation behavior section formed in the graph are as follows: S301: Call the periodic signal sequence corresponding to each fence node in the perturbation stacking section, extract the signal amplitude value and the direction change angle in each period, pair them as equally spaced signal observation points in chronological order, and perform joint division of amplitude and direction based on the time period range to generate a synchronous statistical input fragment set; To call the periodic signal sequences corresponding to each fence node in the perturbed stacked section, it is first necessary to extract the time periods marked as perturbed stacking from the historical monitoring data files of each node. All the sampling period data within this time period are regarded as target periods. From these, the signal amplitude values and direction change angles recorded in each period are extracted one by one. The signal amplitude can be directly calculated from the voltage change value, and the method is the absolute value of the difference between the voltage of this period and the voltage of the previous period. The direction change angle is calculated according to the included angle of spatial vectors. The calculation method of the included angle is to take the dot product of the direction vectors of two periods and then divide by the product of the modulus lengths, and further obtain the angle value by inverse trigonometry. Each time the signal amplitude and direction angle results are extracted, the results are sorted by time and added to a set of equally spaced signal observation points. For example, if a node samples continuously 6 times between the 10th second and the 12th second, and the sampling period is once every 2 seconds, then 3 observation points will be formed. Each observation point contains a set of data of the amplitude and direction angle of this period. Then, according to the time period range where each group of observation points is located, such as 10–12 seconds, 12–14 seconds, etc., all the observation points are divided into these interval segments. All the observation points within each time period form a statistical input segment. The division principle is to perform segmentation based on a fixed length of the time window (such as 2 seconds). If the timestamps in the original record do not meet the neat division, forward filling is used to supplement the observation data to ensure that the number of observation points is the same between all time period segments, thus forming a complete set of synchronized statistical input segments.

[0032] S302: According to the signal amplitude and direction angle data of the observation points in the synchronized statistical input segment set, calculate the signal intensity range, direction switching frequency, and perturbation duration frames of adjacent periods respectively, and normalize and combine the three structural values into a multi-dimensional statistical index to construct a stacked section structure parameter group; According to the signal amplitude and direction angle data of the observation points in the synchronized statistical input segment set, first perform a difference operation on the signal amplitudes between any two consecutive periods to obtain the signal intensity range. This range is the absolute value of the difference between the amplitude of the latter period and the amplitude of the previous period, which is used to describe the perturbation fluctuation. Then, count whether the direction angle change within a certain section exceeds the set angle threshold, such as 30 degrees. If the angle change between two consecutive periods exceeds 30 degrees, it is counted as one direction switch. In this way, the direction switching frequency within the entire segment is statistically obtained. The direction switching frequency is equal to the number of switches divided by the total length of the time period. At the same time, record the number of frames with continuous perturbation signals within this time period. The judgment basis is that only when the amplitude value is greater than the set amplitude threshold is it regarded as a perturbation frame. If the amplitude is continuously greater than 0.8V for three frames in a certain section, the perturbation duration frames in this section are 3. The three structural items are the range value, switching frequency, and perturbation frames. Then, perform normalization operations on the three items of data respectively. After normalization, the three items of data are combined into a three-dimensional vector structure. In the way that each time period corresponds to a three-dimensional index, a stacked section structure parameter group is constructed.

[0033] S303: calling the stacked segment structure parameter group, constructing a two-dimensional density distribution map with the signal strength extreme difference as the horizontal axis and the direction switching frequency as the vertical axis, identifying the area with the maximum statistical density in the map as the disturbance focus block, and outputting the boundary disturbance behavior segment characteristics based on the distribution range of the focus block; When calling the stacked segment structure parameter group, the first dimension representing the extreme difference of signal strength in each three-dimensional vector is used as the horizontal axis, and the second dimension of the direction switching frequency is used as the vertical axis. The parameter coordinates corresponding to each observation point are plotted in the two-dimensional coordinate system to form a two-dimensional scatter plot. Then, by counting the density of points around each coordinate point, the area with the most points is determined as the block with the largest density. The density evaluation method is to set a certain coordinate point as the center and define a neighborhood area with a radius of R, for example, R is 0.1, and count the number of points contained in the area. If the number of points in the neighborhood of a certain point is the largest, it is the area where the density maximum point is located. Then, a boundary enclosing structure is formed for all points in the area to obtain the corresponding disturbance behavior time interval. The boundary formation method can be defined by the maximum and minimum horizontal and vertical coordinate values, or the contour scanning method can be used to establish a closed curve according to the natural boundary of the density. Finally, all observation time periods involved in the boundary area are reviewed to form the output of the boundary disturbance behavior segment characteristics.

[0034] See also Figure 5 , obtain the resistance response value sequence of the fence node corresponding to the boundary disturbance behavior segment feature and perform boundary response behavior recognition. The specific steps of capturing the boundary burst response set are as follows: S401: calling the fence node number corresponding to the boundary disturbance behavior segment feature, obtaining the resistance response value sequence of the fence node in the corresponding cycle, extracting the resistance change interval and the minimum response value in each cycle, and generating a fence node resistance sequence set; To call the fence node number corresponding to the boundary disturbance behavior segment feature, it is necessary to first extract all the included fence node index numbers from the disturbance focus block result output in the previous step. These numbers will be used as the retrieval basis to call out the resistance response records of the corresponding nodes in a specific period from the monitoring database. Each resistance response value is the result of the reverse calculation after the current and voltage changes, which comes from the voltage measured on site divided by the current value. The sampling period is usually set synchronously with the voltage and direction, for example, once every 2 seconds, so the resistance sequence has good time continuity. After obtaining these periodic resistance values, two key parameters need to be extracted from them: one is the resistance change interval, which represents the difference between the maximum resistance and the minimum resistance in a period, reflecting the resistance fluctuation amplitude of the period; the other is the minimum response value, that is, the minimum value of all resistance records in the period, which is used to indicate the weakest state of conduction capability. The two parameters corresponding to all periods are arranged in time to form a resistance sequence set.

[0035] S402: According to the fence node resistance sequence set, calculate the resistance transient amplitude, resistance rebound duration, and jump duration period for each fence node within the calculation period. Respectively using the amplitude difference, fluctuation recovery duration, and abnormal period quantity as calculation items, merge multiple data vectors to construct a fence node resistance behavior parameter group; According to the fence node resistance sequence set, it is necessary to further perform a detailed structural calculation on the resistance change characteristics of each period. First is the resistance transient amplitude, that is, the change amplitude of the minimum resistance between adjacent periods. For example, if the minimum resistance in the previous period is 1.5 and the current period is 2.0, then the transient amplitude is 0.5; the second item is the resistance rebound duration, which refers to the time length experienced by the resistance when rising from a low value to a high value range. The judgment process is the number of periods required for the resistance to continuously rise to near the starting level after a local minimum appears in the continuous monitoring period. For example, period 5 is the low point, and periods 6 to 9 rise to near the high value of period 4, and the rebound duration is 4 periods; the third item is the jump duration period, that is, the number of consecutive periods in which the resistance is higher than the set abnormal threshold or lower than the set extremely low threshold. The threshold setting can refer to the historical baseline resistance fluctuation range. Usually, a high abnormal threshold such as 2.5Ω and a low abnormal threshold such as 1.2Ω are set. If it is lower than 1.2Ω for three consecutive periods, then the jump duration period is 3; represent the results obtained from the above three calculations in a unified structural vector, and group them according to the fence nodes, and finally form a fence node resistance behavior parameter group.

[0036] S403: Invoke the feature vectors of each fence node in the fence node resistance behavior parameter group, perform a vector clustering distribution judgment based on distance similarity, and identify the fence nodes deviating from the normal clustering center through the isolation forest algorithm to screen the boundary burst response set; After invoking the feature vectors of each fence node in the fence node resistance behavior parameter group, it is necessary to first perform a vector clustering distribution judgment based on distance similarity. For this purpose, it is necessary to standardize the three-dimensional feature vectors of all nodes. The feature vector of each fence node contains three dimensions: resistance transient amplitude, resistance rebound duration, and jump duration period. To ensure the comparability of numerical values under different dimensions, it is necessary to uniformly perform processing in the minimum-maximum normalization method. The normalization result of each item is , after all nodes are calculated, the standardized feature vectors are obtained. For example, for the feature vector of a certain fence node, after normalization, it is , and its corresponding explanation is as follows: : The standardized value of the resistance transient amplitude of this node. The original value is the difference between the minimum resistance of the current period and the previous period, reflecting the severity of the response; : The standardized value of the resistance rebound duration of this node. The original value is the number of periods used for the resistance to rise from the minimum value to the initial state multiplied by the period duration, indicating the recovery speed; : The normalized value of the jump duration period of this node. The original value is the number of consecutive periods of the abnormal resistance value multiplied by the period duration, reflecting the continuous abnormal time; the corresponding eigenvector of any other node is , and their meanings are as follows: : The normalized value of the transient amplitude of the resistance of another node; : The normalized value of the resistance rebound time of this node; : The normalized value of the jump duration of this node; to evaluate the behavioral similarity between nodes, the normalized Euclidean distance is used to calculate the distance between two nodes: , and the above distance is used to construct distance matrix (where is the total number of nodes), providing a basis for subsequent cluster structure analysis and density evaluation.

[0037] In the abnormal recognition process after obtaining the node distribution, the recognition of boundary burst behavior is performed through the isolation forest algorithm. This algorithm constructs the following function based on path length analysis combined with local density factors: ; where the explanation of each parameter is as follows: is the abnormal degree value of the th fence node, used to measure whether the target node is an abnormal node deviating from the cluster center. The larger the value, the higher the deviation degree; : The th fence node, corresponding three-dimensional normalized eigenvector; : The total number of samples participating in the calculation, that is, the number of fence nodes; : The input of the harmonic number function, representing the comparison set formed by excluding the current point from the total number of samples; : The approximation of the harmonic number, representing the sum from added to , and the approximation formula is , where is the Euler–Mascheroni constant, used to estimate the depth of the random tree; : The sample partition correction term, representing the penalty term for the average isolation cost in forest construction; : The sample constant term, also the correction constant for the theoretical average path length. The physical meaning of the entire expression is "under samples, the theoretical expected path depth", with the unit of "number of nodes"; if the total number of samples , then: , , then ; : Sample The average path length required to be split and isolated in all isolation forests is automatically calculated by the isolation forest model; : The local density index of the sample is the average Euclidean distance between it and its five nearest neighbors , and then take its reciprocal, that is , with the unit of "reciprocal of distance", used to measure the sparsity of nodes in space; : The local density adjustment coefficient, which controls the influence degree of density on the score. The setting basis is the gradient response of the measured abnormal distribution density curve. In this scenario, it is set to 0.2, which is an empirical value verified by experiments to effectively distinguish the aggregation area from isolated points. If the setting is too small, the density difference is not enough to adjust the abnormal weight; if the setting is too large, the density will dominate and cover up the real path anomaly; Let , , , Then there is: . After statistically calculating the mean and standard deviation of all node scores , assuming the mean is 1.10 and the standard deviation is 0.12, define the abnormal screening threshold as , node is exactly at the boundary and is regarded as a high-risk node. Finally, output the node number and include it in the boundary emergency response set. This set is the fence node that shows isolated characteristics in terms of structural features during the current period.

[0038] Please refer to Figure 6 to obtain the fence nodes corresponding to the boundary emergency response set, and conduct a joint analysis of the resistance behavior parameter group and the disturbance intensity level of the target fence nodes. The specific steps for dividing the fence fault diagnosis area are as follows: S501: Call the fence nodes in the boundary emergency response set, obtain the fence node resistance behavior parameter group and the corresponding disturbance intensity level of the fence nodes in the previous period, and pair the two according to the fence node number to establish a fence node response index combination; After invoking the fence nodes in the call boundary burst response set, it is necessary to retrieve the resistor behavior parameter groups recorded by these nodes during their previous complete disturbance cycle and pair them with the disturbance intensity levels during that cycle. The resistor behavior parameter group includes three data items: resistor transient amplitude, resistor rebound duration, and jump duration period. The disturbance intensity level is classified based on the disturbance amplitude change rate and direction flip frequency calculated previously. The intensity level is usually divided into three levels, namely level 1 (slight), level 2 (medium), and level 3 (severe), and the setting basis is as follows: The historical average value of the disturbance amplitude change rate is set as the reference value. If it exceeds 0.3V / s and the direction frequency exceeds 0.2Hz, it is regarded as level 2. If the amplitude exceeds 0.5V / s and the frequency exceeds 0.35Hz, it belongs to level 3. These thresholds are derived from the statistical samples of the engineering site and correspond to the upper limits of the 99% safe operation range respectively. Each node corresponds to a resistor parameter group and a disturbance intensity value during this cycle, and the two are matched in the data structure according to the node number, and finally a fence node response index combination is formed.

[0039] S502: According to the resistor transient amplitude and disturbance intensity level of each fence node in the node response index combination, arrange them in a spatial grid structure based on the relative coordinate positions of the nodes in the fence layout, perform gradient cross calculations on the grid points and superimpose the mapping values to generate a fence space joint response map; According to the resistor transient amplitude and disturbance intensity level data of each fence node in the node response index combination, first map all the nodes to a two-dimensional spatial grid structure according to their actual coordinates in the physical layout. Each node corresponds to a grid point, and the coordinate system is automatically matched according to the construction wiring sequence or geographical coordinates. For example, the horizontal axis is the column number where the node is located, and the vertical axis is the row number where it is located. After establishing the grid, it is necessary to perform gradient cross calculations on each grid point. The purpose is to analyze whether there is a cross-enhancement phenomenon between the resistor change trend and the disturbance level change direction in space. For this reason, a scalar gradient field is constructed based on two physical fields: one is the resistor transient amplitude field, with the unit of ohm ( ); the other is the disturbance intensity level field, which is a dimensionless discrete value (level 1–3). In the two-dimensional grid, the following improved formula is used to define the response intensity of the grid point: ; where: : represents the gradient cross response value at the grid coordinate , with the unit of , indicating the coupling intensity of the spatial changes of the resistor and the disturbance level within the specified area; : the resistor transient amplitude at this point, with the unit of , from the resistor behavior parameter group of the previous cycle; : the disturbance intensity level at this point, which is a dimensionless discrete value; : the resistor at and the spatial gradient in the direction (unit: : the disturbance level is at and the spatial gradient in the direction (unit:

[0040] Suppose the left and right transient resistances of a certain node are 0.4 Ω and 0.8 Ω respectively, and the distance is 2 meters, then the horizontal gradient is ; the up and down disturbance levels are 2 and 3 respectively, and the distance is 2 meters, and the gradient is . If the vertical resistance gradient is 0.1 , and the horizontal disturbance gradient is 0.3 , then: ; After performing this calculation for all grid nodes, a combined response map of the fence space is generated. In the combined response map of the fence space, the system will normalize the response values of all grid points, calculate their mean and standard deviation, and thus establish a distribution interval of the response intensity. If the response value of a certain grid point exceeds the mean of all response values in the map plus twice the standard deviation, it can be considered that the crossing intensity of this point is significantly higher than the average level and belongs to the "high response area"; if it exceeds the mean plus three times the standard deviation, it is regarded as an extremely high response area. Such areas are usually marked as potential fault conduction hotspots or behavior focus blocks. The setting of this standard comes from the outlier detection principle in statistics, and at the same time combines the on-site layout density and the actually measured disturbance response range, and has been verified to have good engineering applicability. In other words, "high response value" does not refer to a certain absolute value, but refers to the relative position in the overall response field. It is higher than the upper limit of the main distribution interval and is an important signal for abnormal evolution to form cross mutations in space.

[0041] S503: Call all grid points in the combined response map of the fence space, identify the set of fence nodes whose disturbance intensity level exceeds one time the average value of all nodes in the area where the resistance rebound trend slows down, perform attribution determination on the spatial range of the set and output it as the fence fault diagnosis area; After the generation of the combined response map of the fence space is completed, the system will conduct a centralized analysis of all grid points in the map to identify the intersection nodes of the area where the resistance rebound trend slows down and the area where the disturbance intensity level is abnormal. The criterion for judging the slowdown of the resistance rebound trend is that the rebound duration of the node increases in multiple consecutive cycles, and the rebound duration in the latest cycle exceeds twice the historical rebound average of the node; for example, if the historical average rebound time of a node is 6 seconds and the current time is 13 seconds, it is considered that the trend has slowed down. At the same time, the determination of the area with abnormal disturbance intensity level is based on setting a screening threshold according to the average disturbance level of all nodes. For example, if the average level of all nodes is 1.9, the system will extract the grid points of the nodes with a level ≥ 3.8 (i.e., more than double). After the joint screening of these two conditions, the system will mark the set of nodes that meet the above criteria in the map. Then, based on the distribution density and adjacency relationship of the set nodes in space, connectivity analysis and regional aggregation are performed, a closed boundary is delimited, and an identifiable spatial fault block is formed. The regional aggregation method combines graph connectivity clustering and geometric contour fitting: if the distance between adjacent abnormal nodes is less than a certain threshold (for example, 2.5 meters), they are classified into the same block. Finally, the system can obtain the following typical fence fault diagnosis area types according to this scheme (example classification): Isolated response concentration area: High-value response points located at the edge or corner of the fence, usually related to poor contact with the ground grid or end voltage reflection; Strip-shaped abnormal conduction area: A high-response band continuously distributed along the fence, reflecting the propagation of disturbances from one end to the other, and may be related to loose contact of the main conductor or frequent jump areas; Focused disturbance concentration and dispersion area: A relatively concentrated and asymmetric response block area inside the fence, usually indicating local component aging or environmental interference points, such as moisture, grass and wood intrusion, etc.; Grid-shaped scattered area: High-response points distributed discretely in a checkerboard pattern, which may be related to intermittent release of interference sources, uneven cable shielding, or simultaneous degradation at multiple points; These diagnosis areas will be used as the output content of the fault warning and presented through interface marking, coordinate output or interactive layer methods for maintenance personnel to conduct target review and further handling in combination with the on-site layout and sensing node positioning.

[0042] Please refer to Figure 7 , a fault diagnosis system for the safety fence of a power project. The system includes: A disturbance factor identification module, which obtains the conduction voltage and disturbance direction of the fence at multiple time points through the fence conduction fence nodes, conducts a time-series evolution modeling of the disturbance trend term, and identifies the disturbance factor group of the fence nodes; A disturbance aggregation screening module, which judges the aggregation trend of the conduction voltage sequence and screens the corresponding disturbance stacking sections according to the disturbance change conditions of the disturbance factors in the disturbance factor group of the fence nodes; A boundary behavior extraction module, which obtains the periodic signal sequence of the disturbance stacking section, constructs a critical domain active graph and identifies the characteristics of the boundary disturbance behavior sections formed in the graph; The resistance response recognition module obtains the resistance response value sequence of the fence nodes corresponding to the characteristics of the boundary disturbance behavior section and performs boundary response behavior recognition to capture the boundary sudden response set; The fault diagnosis generation module obtains the fence nodes corresponding to the boundary sudden response set, jointly analyzes the resistance behavior parameter group of the target fence nodes and the disturbance intensity level, and divides the fence fault diagnosis area.

[0043] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A fault diagnosis method for the safety fence in electric power engineering, characterized in that, It includes the following steps: S1: Obtain the conduction voltages and disturbance directions of the fence nodes at multiple time points through the conduction of the fence, perform time-series evolution modeling of the disturbance trend terms, and identify the disturbance factor groups of the fence nodes; S2: According to the disturbance change conditions of the disturbance factors in the disturbance factor groups of the fence nodes, judge the aggregation trend of the conduction voltage sequence and screen the corresponding disturbance stacking sections; S3: Obtain the periodic signal sequence of the disturbance stacking section, construct a critical domain activity graph, and identify the characteristics of the boundary disturbance behavior sections formed in the graph; S4: Obtain the resistance response value sequence of the fence nodes corresponding to the characteristics of the boundary disturbance behavior sections and perform boundary response behavior identification to capture the boundary sudden response set; S5: Obtain the fence nodes corresponding to the boundary sudden response set, jointly analyze the resistance behavior parameter groups and disturbance intensity levels of the target fence nodes, and divide the fence fault diagnosis area.

2. The power engineering safety fence fault diagnosis method according to claim 1, characterized in that: The disturbance factor groups of the fence nodes are specifically the disturbance amplitude change rate, the disturbance direction flip frequency, and the disturbance trend slope. The disturbance stacking sections include the disturbance density peak section, the disturbance amplitude superposition interval, and the disturbance period concentration area. The characteristics of the boundary disturbance behavior sections include the signal intensity range area, the direction switching concentration section, and the disturbance duration frame section. The boundary sudden response set is specifically the abnormal resistance jump fence nodes, the non-stable resistance rebound fence nodes, and the resistance jump continuous abnormal fence nodes. The fence fault diagnosis area includes the fault boundary fence node group, the diagnostic space position partition, and the joint feature concentrated distribution area.

3. The power engineering safety fence fault diagnosis method according to claim 1, characterized in that: The specific steps for obtaining the conduction voltages and disturbance directions of the fence nodes at multiple time points through the conduction of the fence, performing time-series evolution modeling of the disturbance trend terms, and identifying the disturbance factor groups of the fence nodes are as follows: S101: Obtain the measured conduction voltage values, current change rates, and signal direction vectors of each fence conduction fence node at the power engineering site during consecutive sampling periods. Based on the measured conduction voltage values, construct a conduction voltage sequence according to the sampling order. Based on the signal direction vectors, calculate the direction change angle sequence between adjacent periods to generate a voltage and direction input data group; S102: Based on the voltage and direction input data group, calculate the periodic disturbance amplitude change rate for the conduction voltage sequence, judge the number of direction changes within a specified time for the direction change angle sequence, and then pair and recombine the two into a continuous sequence structure to generate a disturbance trend modeling input vector set; S103: Perform continuous periodic input processing on the data pairs of each fence node in the disturbance trend modeling input vector set, use the autoregressive moving average algorithm to perform time trend sequence modeling, extract the disturbance amplitude change rate and disturbance direction flip frequency of each group of fence nodes in each period, and obtain the disturbance factor groups of the fence nodes.

4. The fault diagnosis method for the safety fence in the power project according to claim 1, characterized in that: The specific steps for judging the aggregation trend of the conduction voltage sequence and screening the corresponding disturbance stacking sections according to the disturbance change conditions of the disturbance factors in the disturbance factor groups of the fence nodes are as follows: S201: Call the perturbation amplitude change rate and perturbation direction flip frequency corresponding to each fence node in the fence node perturbation factor group, and combine the periodic arrangement structure of the same fence node in the conduction voltage sequence to construct a periodic signal sequence set; S202: According to the signal amplitude of each fence node in the periodic signal sequence set, perform periodic superposition on the continuous periodic perturbation amplitudes, calculate the perturbation amplitude accumulation of each periodic fence node, arrange them in the sampling order to form a perturbation accumulation curve, perform density statistics on the curve, and calibrate the perturbation density peak interval; S203: Call the perturbation amplitude accumulation rate of the corresponding fence node in the perturbation density peak interval, and perform a ratio judgment with the average value of the perturbation amplitude change rate in the fence node perturbation factor group. If the former is twice or more than the latter, screen the corresponding time period and output it as the perturbation stacking section.

5. The power engineering safety fence fault diagnosis method according to claim 1, characterized in that: The specific steps to obtain the periodic signal sequence of the perturbation stacking section, construct the critical domain activity map, and identify the characteristics of the boundary perturbation behavior section formed in the map are as follows: S301: Call the periodic signal sequence corresponding to each fence node in the perturbation stacking section, extract the signal amplitude value and direction change angle in each period, pair them as equally spaced signal observation points in chronological order, and perform joint division of amplitude and direction based on the time period range to generate a synchronous statistical input segment set; S302: According to the signal amplitude and direction angle data of the observation points in the synchronous statistical input segment set, calculate the signal intensity range, direction switching frequency, and perturbation duration frames of adjacent periods respectively, and normalize and combine the three structural values into a multi-dimensional statistical index to construct a stacking section structure parameter group; S303: Call the stacking section structure parameter group, construct a two-dimensional density distribution map with the signal intensity range as the horizontal axis and the direction switching frequency as the vertical axis, identify the area with the largest statistical density in the map as the perturbation focus block, and output the characteristics of the boundary perturbation behavior section based on the distribution range of the focus block.

6. The power engineering safety fence fault diagnosis method according to claim 1, characterized in that: The specific steps to obtain the resistance response value sequence of the fence node corresponding to the characteristics of the boundary perturbation behavior section and perform boundary response behavior recognition, and capture the boundary burst response set are as follows: S401: Call the fence node number corresponding to the characteristics of the boundary perturbation behavior section, obtain the resistance response value sequence of the fence node in the corresponding period, extract the resistance change interval and the minimum response value in each period, and generate a fence node resistance sequence set; S402: According to the fence node resistance sequence set, calculate the resistance transient amplitude, resistance rebound duration, and jump duration period of each fence node in the period, use the amplitude difference, fluctuation recovery duration, and abnormal period number as calculation items respectively, combine multiple data vectors, and construct a fence node resistance behavior parameter group; S403: Call the feature vector of each fence node in the fence node resistance behavior parameter group, perform a vector clustering distribution judgment based on the distance similarity, and identify the fence nodes deviating from the normal clustering center through the isolation forest algorithm to screen the boundary burst response set.

7. The power engineering safety fence fault diagnosis method according to claim 6, characterized in that: For the fence nodes identified by the isolation forest algorithm as deviating from the normal clustering center, use the formula: ； Calculate the anomaly degree value of the th fence node. After calculating the anomaly degree values of all fence nodes, calculate the mean and standard deviation of the overall anomaly degree value. If the anomaly degree value of a node exceeds the mean plus twice the standard deviation, it is a fence node that deviates from the normal clustering center; ​ Among them, is the th fence node, is the sample constant term, reflecting the average path length correction amount when the sample size in the isolation forest is . is the average path length required for the sample to be split and isolated in all isolation forests, is the local density index of the sample , is the local density adjustment coefficient, used to control the influence of the density term on the anomaly degree value.

8. The power engineering safety fence fault diagnosis method according to claim 1, characterized in that: Obtain the fence nodes corresponding to the boundary burst response set, and conduct a joint analysis of the resistance behavior parameter group of the target fence nodes and the disturbance intensity level. The specific steps for dividing the fence fault diagnosis area are as follows: S501: Call the fence nodes in the boundary burst response set, obtain the resistance behavior parameter group of the fence nodes and the corresponding disturbance intensity level in the previous cycle of the fence nodes, pair the two according to the fence node numbers for data, and establish a fence node response index combination; S502: According to the resistance transient amplitude and disturbance intensity level of each fence node in the node response index combination, arrange them in a spatial grid structure based on the relative coordinate positions of the nodes in the fence layout, perform gradient cross calculations on the grid points and superimpose the mapping values to generate a fence space joint response map; S503: Call all the grid points in the fence space joint response map, identify the set of fence nodes whose disturbance intensity level exceeds one time the average value of all nodes in the area where the resistance rebound trend slows down, perform attribution determination on the spatial range of the set and output it as the fence fault diagnosis area.

9. The power engineering safety fence fault diagnosis method according to claim 8, characterized in that: Perform gradient cross calculations on the grid points, using the formula: ； Wherein: represents the gradient cross response value at which indicates the coupling strength of the spatial variation of the resistance and the perturbation level within the specified area, is the resistance transient amplitude of the target point, is the perturbation intensity level of the target point, is the resistance at and the spatial gradient in the direction of, is the perturbation level at and the spatial gradient in the direction of.

10. A fault diagnosis system for the safety fence in electric power engineering, characterized in that, Execute according to the power engineering safety fence fault diagnosis method described in any one of claims 1-9. The system includes: A disturbance factor identification module, which obtains the conduction voltage and disturbance direction of the fence at multiple time points through the fence-conducting fence nodes, conducts a time-series evolution modeling of the disturbance trend term, and identifies the fence node disturbance factor group; A disturbance aggregation screening module, which judges the aggregation trend of the conduction voltage sequence and screens the corresponding disturbance stacking sections according to the disturbance change situation of the disturbance factors in the fence node disturbance factor group; A boundary behavior extraction module, which obtains the periodic signal sequence of the disturbance stacking section, constructs a critical domain activity map and identifies the characteristics of the boundary disturbance behavior section formed in the map; A resistance response identification module, which obtains the resistance response value sequence of the fence nodes corresponding to the characteristics of the boundary disturbance behavior section and performs boundary response behavior identification to capture the boundary burst response set; A fault diagnosis generation module, which obtains the fence nodes corresponding to the boundary burst response set, conducts a joint analysis of the resistance behavior parameter group of the target fence nodes and the disturbance intensity level, and divides the fence fault diagnosis area.

Images