A method for real-time analysis and monitoring of connection state of ground down conductor

Abstract

The present application relates to the technical field of grounding downwire connection state monitoring, and particularly relates to a grounding downwire connection state real-time analysis monitoring method, which comprises the following steps: real-time acquisition of multi-dimensional monitoring data to construct a characteristic vector, extraction of outlier timestamp sequences and amplitude deviation in a window to calculate time sequence pulse characteristic values of each dimension; translation comparison of sequences of a reference dimension and a related dimension, calculation of an outlier coupling correction factor in combination with characteristic differences; acquisition of noise confidence according to the foregoing, and statistics of global noise interference strength; finally, dynamic adjustment of a neighborhood radius and a minimum sample number of a spatial clustering model by using the interference strength, clustering evaluation of the characteristic vector, and triggering of a warning. The present application distinguishes environmental noise through cross-coupling analysis of multi-dimensional time sequence pulses, and adaptively adjusts clustering model parameters, thereby solving the problem of false positives and false negatives in state monitoring under complex environments, and improving the accuracy of grounding downwire connection state monitoring.

Description

Technical Field

[0001] This invention relates to the field of grounding down conductor connection status monitoring technology, and in particular to a method for real-time analysis and monitoring of grounding down conductor connection status. Background Technology

[0002] The grounding down conductor of the tower is a key channel for lightning protection and fault current discharge of transmission lines. The reliability of its connection status directly determines the operational safety of the power grid system. In actual operation and maintenance, mechanical loosening or electrochemical corrosion of the down conductor is often accompanied by the coordinated changes of multiple physical parameters. Therefore, it is usually necessary to conduct real-time joint monitoring of multi-dimensional characteristics such as grounding resistance, leakage current and soil resistivity in order to achieve early warning of hidden faults.

[0003] For the collected multidimensional monitoring data, existing technologies generally use the unsupervised DBSCAN clustering algorithm to build a state recognition model. The method is mainly based on the density distribution of data in the feature space. It defines the density threshold of clustering by pre-setting two core parameters: a fixed neighborhood radius and a minimum number of samples. Then, the data samples are divided into fault points, boundary points, or noise points to capture the feature differences under different connection states.

[0004] However, in complex field operating environments, when encountering strong electromagnetic interference or severe thunderstorms causing drastic fluctuations in circuit impedance, monitoring data is prone to generating large-scale transient scattered noise points. The limitation of traditional monitoring methods is that their clustering parameters are static and fixed. When a large number of randomly approaching environmental noise points emerge, they are very likely to reach the statically set density threshold, thus being incorrectly linked into false fault clusters by the algorithm. This causes the system to misjudge random environmental noise as real physical connection degradation, not only causing frequent invalid false alarms, but also seriously masking the real weak degradation characteristics, thus losing the reliability of condition monitoring and early warning. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a real-time analysis and monitoring method for the connection status of grounding down conductors, in order to solve the problem that existing status monitoring methods based on fixed parameter clustering are prone to misclustering large-scale transient environmental noise into false fault clusters when facing complex field electromagnetic and meteorological interference, resulting in high false alarm rates and unreliable fault identification of traditional monitoring models.

[0006] This invention provides a method for real-time analysis and monitoring of the connection status of grounding down conductors, comprising the following steps: acquiring standardized multi-dimensional data of grounding down conductors in real time and constructing a real-time feature vector; dividing the current window based on the current time, acquiring outliers in each dimension of the data within the current window, and constructing a timestamp sequence of outliers for each dimension in chronological order; calculating the time-series pulse characteristic value of each dimension based on the interval between adjacent timestamps in the outlier timestamp sequence and the maximum amplitude deviation of the data segments before and after the outlier; selecting a reference dimension from the multi-dimensional data, and using the remaining dimensions as correlated dimensions; and performing a translational comparison between the timestamp sequence of outliers in the reference dimension and any correlated dimension to obtain synchronization. The number of outliers is used to calculate the outlier coupling correction factor based on the difference in time-series pulse feature values ​​between the two. The time-series pulse feature values ​​of the baseline dimension are weighted based on the minimum value of the outlier coupling correction factor to obtain the noise confidence level, and noise salient windows are distinguished based on the empirical classification threshold. The proportion of noise salient windows within a preset historical time period is statistically analyzed to obtain the global noise interference intensity. The initial preset neighborhood radius and the initial preset minimum sample number of the spatial clustering model are dynamically adjusted using the global noise interference intensity. The spatial clustering model with adjusted parameters is used to cluster the real-time feature vectors. The abnormal probability of the current feature vectors is evaluated based on the clustering results to achieve real-time early warning of the grounding down conductor connection status.

[0007] Preferably, the step of obtaining outliers in each dimension of data within the current window and constructing an outlier timestamp sequence for each dimension in chronological order includes: using a local anomaly factor algorithm to detect outliers in each dimension of data within the current window and identifying outliers in each dimension; extracting the timestamps corresponding to the outliers and arranging them in ascending order according to chronological order to obtain an ordered outlier timestamp sequence for each dimension.

[0008] Preferably, the method for obtaining the maximum amplitude deviation of the data segments before and after the outlier is as follows: if the number of outliers in a certain dimension of data within the current window is less than 2, then the time-series pulse feature value of that dimension is directly set to 0; if the number of outliers in a certain dimension of data within the current window is greater than or equal to 2, then the time-series data segments before and after the timestamp corresponding to each outlier are extracted respectively, and the first mean value of the monitored values ​​of the previous time-series data segment and the second mean value of the monitored values ​​of the subsequent time-series data segment are calculated; the absolute difference between the first mean value and the second mean value is taken as the amplitude deviation of the outlier, and the maximum value of the amplitude deviation of all outliers in that dimension within the current window is extracted as the maximum amplitude deviation.

[0009] Preferably, the calculation of the time-series pulse feature values ​​for each dimension includes: calculating the average value of all adjacent timestamp intervals in the outlier timestamp sequence for each dimension, and normalizing the average value to obtain a first feature; normalizing the maximum amplitude deviation, and subtracting the normalized maximum amplitude deviation from a constant 1 to obtain a second feature; and multiplying the first feature and the second feature to obtain the time-series pulse feature values ​​for each dimension.

[0010] Preferably, the step of translating and comparing the outlier timestamp sequences of the baseline dimension and any associated dimension to obtain the number of synchronized outliers includes: keeping the element index boundaries of the outlier timestamp sequences of the baseline dimension unchanged; traversing and translating the outlier timestamp sequences of any associated dimension in the direction of increasing or decreasing index by a set translation step size; under each translation amount, extracting the valid point pairs within the index overlap interval of the outlier timestamp sequences of the baseline dimension and the translated outlier timestamp sequences of any associated dimension, and calculating the average absolute difference of the timestamps of all valid point pairs as the time difference; determining the translation amount corresponding to the minimum time difference as the optimal translation amount, and extracting the total number of valid point pairs under the optimal translation amount, denoted as the number of synchronized outliers.

[0011] Preferably, the calculation of the outlier coupling correction factor includes: calculating the ratio of the number of synchronized outliers to the total number of outliers in the baseline dimension; calculating the absolute value of the difference between the temporal pulse feature value of the baseline dimension and the temporal pulse feature value of any associated dimension, performing an exponential operation with the natural constant as the base and the opposite of the absolute value as the exponent to obtain a feature difference attenuation term; and multiplying the ratio by the feature difference attenuation term to obtain the outlier coupling correction factor between the baseline dimension and any associated dimension.

[0012] Preferably, the step of obtaining the noise confidence score and distinguishing noise salient windows based on an empirical classification threshold includes: extracting the minimum value among the outlier coupling correction factors corresponding to the baseline dimension and all associated dimensions within the current window, denoted as the comprehensive coupling correction factor; and calculating the noise confidence score of the current window. , In the formula, Represents the temporal pulse characteristic value of the baseline dimension within the current window; Indicates the comprehensive coupling correction factor; when When the noise level exceeds the empirical classification threshold, the current window is marked as a noisy window.

[0013] Preferably, the step of dynamically adjusting the initial preset neighborhood radius and the initial preset minimum sample number of the spatial clustering model using the global noise interference intensity includes: calculating the adjusted minimum sample number parameter. , In the formula, The initial minimum number of samples is preset; Given the global noise interference intensity; calculate the adjusted neighborhood radius. , In the formula, This is the initial preset neighborhood radius.

[0014] Preferably, the step of clustering real-time feature vectors using a spatial clustering model with adjusted parameters, and evaluating the anomaly probability of the current feature vector based on the clustering results to achieve real-time early warning of the grounding conductor connection status, includes: the spatial clustering model employing the density-based spatial clustering algorithm DBSCAN; after clustering the real-time feature vectors, extracting the Mahalanobis distance between the current feature vector and the center of its core cluster as the spatial deviation of the current feature vector; and normalizing the spatial deviation to a negative exponential mapping function. Within the interval, the anomaly probability of the current feature vector is obtained; when the anomaly probability exceeds the preset multi-level security threshold, the corresponding warning instruction is triggered.

[0015] Preferably, the method further includes: the multi-dimensional data specifically includes: grounding resistance, leakage current and soil resistivity; wherein, the baseline dimension is grounding resistance; and the associated dimensions include leakage current and soil resistivity.

[0016] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: This invention first analyzes the degree of temporal isolation and multi-dimensional system linkage of fluctuations by calculating the time-series pulse characteristic value and outlier coupling correction factor. This mechanism effectively distinguishes between independent and disordered random environmental noise caused by lightning and other factors, and real physical degradation caused by metal rusting and loosening. Second, it uses the assessed global noise interference intensity as a feedback signal to actively intervene in the underlying mechanism of the clustering algorithm, automatically narrowing the detection range in noisy environments to suppress false alarms. Under normal operating conditions, it can automatically smooth regression, maintaining the sensitivity to identify very early and subtle degradation. Finally, based on the adaptive clustering results, it transforms spatial deviation into anomaly probability and combines it with multi-level safety thresholds to trigger graded early warning commands, breaking the limitations of traditional qualitative judgment and providing a reliable basis for cascade condition-based maintenance and emergency repair decisions in power fields. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a flowchart of a method for real-time analysis and monitoring of the connection status of a grounding lead wire provided in Embodiment 1 of the present invention. Detailed Implementation

[0019] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0020] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0021] This invention provides a method for real-time analysis and monitoring of the connection status of grounding down conductors, such as... Figure 1 As shown, the method includes the following steps: Step S101: Obtain standardized multi-dimensional data of the grounding down conductor and construct a real-time feature vector.

[0022] It should be noted that in the actual operation of power poles, the grounding down conductor is not only affected by its own physical state, but also by a complex electromagnetic and meteorological environment in real time. On the one hand, the connection point between the down conductor and the grounding grid, due to long-term burial, is affected by electrochemical corrosion or mechanical loosening, which can lead to a slow and gradual shift in grounding resistance. On the other hand, lightning strikes and line induced overvoltages can cause instantaneous high-frequency pulses to be generated by sensors, and meteorological factors such as rainfall leading to a surge in soil moisture and temperature fluctuations can directly drive significant fluctuations in soil resistivity. In order to comprehensively analyze its connection status and isolate environmental interference, a single data source is often limited. Therefore, it is necessary to collect multi-source heterogeneous time-series data covering electrical and environmental characteristics. Among them, electrical characteristics are used to capture the direct physical feedback of the connection status, while environmental characteristics serve as a reference to isolate natural interference. This multi-dimensional data can provide basic data support for subsequent analysis of systematic linkage changes or independent random disturbances between different indicators, thereby achieving accurate determination of hidden faults.

[0023] Specifically, for the grounding down conductor of the target tower, a multi-dimensional state sensing sensor group is deployed at the connection between the grounding down conductor and the grounding grid, near the tower foundation and the surrounding soil area, to continuously collect time-series data of grounding resistance, leakage current and soil resistivity according to a preset acquisition frequency.

[0024] It should be added that the sampling frequency is used to control the temporal granularity of monitoring data acquisition. Its value is based on a rigorous derivation of the Nyquist sampling theorem and the physical signal frequency band: In the field, weather changes are transient, such as the time constant of soil moisture changes caused by rainfall, which is typically 15 to 30 minutes. According to signal processing theory, in order to effectively preserve and reconstruct a time-duration value on the discrete time axis... Transient physical processes, sampling interval Must meet To address the rapidly changing weather conditions in the field, the sampling interval must be less than 7.5 minutes. This embodiment uses... The 5-minute interval means that 3 to 6 points can be captured continuously within a rainfall cycle, giving it a continuous trend in time. At the same time, for strong electromagnetic pulses such as lightning strikes that are in the millisecond to second range, the 5-minute interval ensures that only extremely isolated single-point mutations are left. The implementers can adjust the interval according to the actual meteorological fluctuation cycle around the substation.

[0025] For the collected raw data, linear interpolation was used to fill in missing records due to temporary sensor failures, communication packet loss, etc. Subsequently, Z-score normalization was performed on the time-series data of all monitoring indicators to eliminate the influence of differences in units between different indicators. The processed time-series data of each indicator were aligned by timestamp. For each moment, a feature vector containing grounding resistance, leakage current, and soil resistivity was constructed. The feature vectors of all moments were arranged in chronological order to form a complete historical monitoring dataset.

[0026] Thus, the standardized multi-dimensional time series dataset was obtained.

[0027] Step S102: Calculate the temporal pulse feature value for each dimension within the current window.

[0028] It should be noted that when lightning strikes, strong winds causing induced overvoltages in the lines, or switching operations occur nearby, the sensors will capture instantaneous signal jumps. This type of environmental noise is extremely sparse on the time axis, and since the noise does not damage the metal connection structure between the down conductor and the grounding grid, the monitoring values ​​such as grounding resistance will quickly return to the original stable baseline after the electromagnetic disturbance ends. In contrast, if the down conductor connection part has electrochemical corrosion due to long-term burial or mechanical loosening due to micro-vibration of the tower base, the conductivity of the contact surface will be substantially damaged, and its monitoring values ​​will show a continuous and irreversible trend shift. If the monitoring data is directly clustered, the system is very likely to mistake the instantaneous jumps that occur frequently during the rainy season for the beginning of a fault. Therefore, by analyzing the degree of isolation of the abnormal fluctuations in single-dimensional data in time, and whether the baseline has undergone substantial displacement after the fluctuation, the pulse attributes of the single-dimensional data fluctuations can be captured, providing a core basis for subsequent removal of random environmental noise.

[0029] Specifically, based on the standardized multi-dimensional time series dataset, the current window is obtained by backtracking a fixed number of sampling times from the past, using the current time as the benchmark, and the data of each dimension within the current window is extracted.

[0030] It should be added that a fixed number of historical backtrackings are used to limit the temporal span of local feature extraction, with values ​​ranging from positive integers. In this invention, 2016 sampling times are used. The logic behind this value is that, at a set frequency of sampling once every 5 minutes, 12 data points are generated per hour, and 12 × 24 × 7 = 2016 corresponds exactly to a complete one-week operating cycle. The core indicators of the grounding down conductor are greatly affected by shallow soil moisture and temperature. A typical natural meteorological process or a complete power grid industrial, agricultural, and commercial load induced fluctuation cycle is usually on a weekly rhythm. A total of 2016 points are used as the analysis window, which can provide a sufficiently rich sample to smooth out the slow seasonal drift, and avoid the window being too small, which would cause the slow change in rainfall over several days to be misjudged as a sudden change. The implementers can adjust this according to the meteorological intensity of the actual substation location and the power grid dispatch cycle requirements.

[0031] Considering that grounding indicators are affected by seasonal factors such as shallow soil temperature and humidity, resulting in slow global background drift, this step introduces a local anomaly factor algorithm to detect outliers in each dimension of data within the current window. By comparing the density difference between the monitored values ​​and their local temporal neighborhoods, the local anomaly factor algorithm can sensitively identify abrupt changes that deviate from the local background, effectively avoiding calculation errors caused by slow environmental drift. Based on this algorithm, the set of timestamps corresponding to the outliers in each dimension of data within the current window is obtained, and the set of timestamps is arranged in ascending order according to time sequence to obtain an ordered sequence of outlier timestamps and the monitoring values ​​corresponding to each outlier.

[0032] For each dimension of data within the current window, the detection logic is executed first: If the number of outliers detected If this indicates that the dimension is extremely stable within the current window and cannot form a basis for pulse feature analysis, then the time-series pulse feature value of this dimension is directly set to 0.

[0033] like Then, deep feature analysis is performed on the data of each dimension within the current window: Obtain the preceding and following time-series data segments of each outlier point in the current window's data dimension. Calculate the mean of the monitoring values ​​of the preceding time-series data segments for each outlier point's timestamp as the first mean. Similarly, calculate the mean of the monitoring values ​​of the following time-series data segments for each outlier point's timestamp as the second mean. The absolute difference between the first and second means is taken as the amplitude deviation of each outlier point. Subsequently, iterate through the amplitude deviations of all outliers in the data dimension and extract the maximum value as the maximum amplitude deviation of the data dimension in the current window.

[0034] Among them, the maximum amplitude deviation is used to capture the most extreme amplitude change event caused by instantaneous strong disturbance within the current window, reflecting whether the shock caused a trend displacement of the physical baseline.

[0035] Based on the interval between adjacent timestamps and the maximum amplitude deviation in the ordered outlier timestamp sequence of each dimension within the current window, the temporal pulse characteristic value of each dimension within the current window is calculated; the specific calculation formula is as follows:

[0036] In the formula, Indicates the number of elements in the current window. Dimensional temporal pulse eigenvalues; Indicates the number of elements in the current window. In an ordered sequence of outlier timestamps of dimension 1 The interval between adjacent timestamps; , These represent the first and second elements within the current window. In a dimensional ordered sequence of outlier timestamps, the index and total number of intervals between adjacent timestamps; Indicates the number of elements in the current window. Maximum magnitude deviation of dimensional data; This represents the maximum and minimum normalization function, used to normalize the calculation results.

[0037] in, Reflects the current window's first The sparsity of the average distribution of outliers in the dimensional data over the time axis; the larger the value, the sparser the distribution of outliers within the current window. The longer and more isolated the time interval of the abnormal jumps in the dimensional data, the more consistent it is with the physical characteristics of lightning side strikes or instantaneous electromagnetic pulses that are occasional and fleeting. For the current window The baseline stability mapping term for dimensional data, if The smaller the value, the more likely it is to be the first element in the current window. After being subjected to an external electromagnetic shock, the grounding resistance or leakage current of the sensor from the dimensional data rapidly returned to the steady-state reference before the jump, which proves that the metal structure of the down conductor did not suffer substantial fracture or oxidation of the contact surface; in summary, by multiplying and weighting the two items, we obtain The larger the value, the higher the value within the current window. The more dramatic the fluctuations in dimensional data, the more likely they are to be independent, disordered, and trendless environmental noise caused by thunderstorms or switching operations, thus distinguishing them from real faults and deterioration caused by rust or loosening.

[0038] At this point, the temporal pulse feature values ​​for each dimension within the current window have been obtained.

[0039] Step S103: Calculate the outlier coupling correction factor between the baseline dimension and any associated dimension within the current window.

[0040] It should be noted that the operating state of a grounding system is characterized by multiple electrical and environmental indicators. True physical connection degradation exhibits strong multi-dimensional systemic linkage characteristics. However, due to the response lag of different sensor hardware or the slight asynchrony in communication sampling, multi-dimensional abnormal jumps caused by the same physical fault often show misalignment in the timestamps recorded by the system. If only absolute timestamps are relied upon for comparison, it is easy to miss true systemic linkages. Therefore, an outlier pair sequence shifting and matching mechanism is introduced. By performing a sequence-level overall shift on the multi-dimensional outlier sequence, the minimum average level of the timestamp difference between two dimensional anomalies is found. If, within a reasonable sequence shift range, the anomaly event sequences of the baseline dimension and the associated dimension are highly aligned, it indicates that they possess stable physical cause consistency and belong to a systemic state evolution; otherwise, they belong to unrelated random noise.

[0041] Specifically, since grounding resistance is the most direct indicator of abnormal connection status, this embodiment selects grounding resistance as the baseline dimension and uses the other dimensions as associated dimensions. That is, a two-dimensional coupling analysis is performed on the baseline dimension and any associated dimension within the current window. While maintaining the element index boundaries of the outlier timestamp sequence in the baseline dimension, the outlier timestamp sequence in any associated dimension is traversed and shifted in the direction of increasing or decreasing index by a set shift step size; the shift step size is set to 1 index unit; under each shift amount, the indexes of the outlier timestamp sequence in the baseline dimension and any associated dimension after shift will overlap, all element pairs within this overlap interval are extracted to form valid point pairs, and the average of the absolute differences of the corresponding timestamps of all valid point pairs is calculated as the time difference for each shift amount; the time differences of all shift amounts within the set allowable range are traversed, and the shift amount with the smallest time difference is determined as the optimal shift amount; the total number of all valid point pairs under the optimal shift amount is extracted and denoted as the number of synchronized outliers.

[0042] For example, when the translation amount is 1, the extracted valid point pairs are: the first point of the outlier timestamp sequence in the baseline dimension is paired with the second point of the outlier timestamp sequence in any associated dimension, the second point of the outlier timestamp sequence in the baseline dimension is paired with the third point of the outlier timestamp sequence in any associated dimension, and so on.

[0043] It should be added that the allowable range is used to limit the maximum sequence misalignment search boundary when pairing outlier sequences in two dimensions. In this invention, it is set to not exceed half of the total number of minimum outliers in the current window to avoid excessive translation that would result in too few matched outlier pairs and lose the true meaning of physical statistics. Implementers can adjust this according to the asynchronous misalignment requirements of the actual multi-source sensor network.

[0044] Based on the difference between the temporal pulse eigenvalues ​​of the baseline dimension and any associated dimension within the current window, and the proportion of synchronous outliers, calculate the outlier coupling correction factor between the baseline dimension and any associated dimension within the current window; the specific calculation formula is as follows:

[0045] In the formula, This indicates the baseline dimension and any associated dimension within the current window. Outlier coupling correction factor; This indicates the total number of outliers in the baseline dimension within the current window; This indicates the baseline dimension and any associated dimension within the current window. The number of synchronous outliers; Represents the temporal pulse characteristic value of the baseline dimension within the current window; Represents any related dimension The time-series pulse characteristic values; This represents the natural exponential function.

[0046] in, This reflects the relationship between the baseline dimension and any associated dimension within the current window. The higher the multidimensional anomaly triggering synchronization rate after excluding hardware asynchronous delay, the stronger the correlation between the abnormal events sensed by different sensors in the sequence occurrence logic. As a nonlinear mapping penalty term for the physical property fit, The smaller the value, the closer the baseline dimension is to any related dimension. The abnormal fluctuations exhibit a highly consistent impulsiveness or trend in their temporal distribution, meaning that the underlying physical causes they face, such as transient lightning strikes or persistent loosening, tend to be more consistent. In this case, the value of the exponential mapping term approaches 1, and no attenuation is applied to the coupling degree; conversely, if... The larger the value, the more independent the fluctuations in the two dimensions are, and the index term will approach 0 to forcibly punish false synchronization; in summary, The larger the value, the closer the baseline dimension is to any associated dimension. The high alignment in the event sequence and the high consistency in the micro-pulse characteristics indicate that the current fluctuations exhibit significant systemic linkage characteristics, which is consistent with the multi-indicator coordinated change law caused by the deterioration of the physical connection status of the grounding down conductor.

[0047] At this point, the outlier coupling correction factor for the baseline dimension and any associated dimension within the current window has been obtained.

[0048] Step S104: Obtain the noise confidence of the current window to evaluate the global noise interference intensity, and calculate the adjusted minimum sample number parameter and neighborhood radius.

[0049] It should be noted that real physical degradation will trigger a systematic linkage of multi-dimensional indicators, while random disturbances usually lack global coordination. This step extracts the minimum coupling correction factor between the baseline dimension and all related dimensions to measure the weakest global correlation. If the minimum value is extremely low, it indicates that there is independent decoupling between indicators, which can further verify the possibility that the current fluctuation is isolated random noise. This global verification mechanism based on the minimum coupling feature suppresses false linkage misjudgments caused by accidental overlap of local dimensions.

[0050] Specifically, the minimum value among the outlier coupling correction factors of the baseline dimension and all related dimensions within the current window is extracted and used as the comprehensive coupling correction factor for the current window.

[0051] The noise confidence level of the current window is obtained by weighting the temporal impulse feature values ​​of the baseline dimension using the comprehensive coupling correction factor of the current window; the specific calculation formula is as follows:

[0052] In the formula, Indicates the noise confidence level of the current window; Represents the temporal pulse characteristic value of the baseline dimension within the current window; This represents the comprehensive coupling correction factor.

[0053] in, As a penalty for multi-dimensional systemic linkage, the larger this value is, the more it indicates that the abnormal mutation of the baseline dimension within the current window lacks real physical coordination with other related dimensions, and the fluctuations of each dimension show a decoupled state of mutual isolation. The larger the value, the more likely the baseline dimension within the window has the dual characteristics of sparse pulse distribution and highly independent multi-dimensional outliers. This means that the data in this period is highly likely to be affected by strong environmental interference and random noise from multiple sensors. Conversely, the smaller the value, the more likely the data in the window has the characteristics of dense single-dimensional outliers and consistent outlier patterns in multiple dimensions. This is more consistent with the system trend changes caused by abnormal actual connection status of the grounding down conductor.

[0054] Furthermore, after obtaining the noise confidence level of the current window, it needs to be transformed into a clear classification state to analyze the severity of the global environment, and an empirical classification threshold is preset. As a decision boundary: when When the current window is determined to have significant noise interference, it is marked as a window with significant noise; when At that time, it is determined that there is no obvious noise interference in the current window, and the data fluctuation mainly comes from changes in the actual state.

[0055] It should be added that the empirical classification threshold is used as the decision boundary for classifying noise interference states. Since the noise confidence value ranges from (0,1), the empirical classification threshold is set to 0.5 in this invention. This value serves as an equivalent equilibrium point in the probability space. Its basis is that it can strictly balance the false alarm rate of the system against occasional strong interference and the false alarm rate against early minor physical degradation in the context of complex substations. The implementers can adjust it based on the actual environment of the substation.

[0056] It should be further explained that if the DBSCAN algorithm parameters are fixed, then when the tower enters the severe rainy season, the large number of electromagnetic and meteorological transient disturbances that emerge in a short period of time will cause the background noise density in the feature space to rise sharply. These scattered discrete noise points generated on a large scale due to environmental degradation are very likely to cluster in the feature space, thereby reaching the statically set clustering threshold and being mistakenly connected by the algorithm into false fault clusters, resulting in a large number of false alarms. Therefore, this step counts the global noise interference intensity and uses it as a feedback signal to actively intervene in the underlying mechanism of the clustering algorithm to ensure that only real physical fault feature points can be identified and clustered.

[0057] Specifically, the total number of windows at the current moment and within the past day is counted, as well as the total number of windows marked as noise significant; the ratio of the total number of windows marked as noise significant to the total number of all windows is calculated as the global noise interference intensity.

[0058] The initial preset minimum sample size parameter and neighborhood radius were adjusted using the global noise interference intensity to obtain the adjusted minimum sample size parameter and neighborhood radius; the specific calculation formulas are as follows:

[0059]

[0060] In the formula, The adjusted neighborhood radius; The initial preset neighborhood radius; This is the adjusted minimum sample size; The initial minimum number of samples is preset; This represents the global noise interference intensity.

[0061] in, This reflects the intensity of global noise interference. The greater the intensity of global noise interference, the smaller the adjusted neighborhood radius, which shrinks the detection range of the clustering algorithm in the local feature space and avoids including scattered noise points in the same neighborhood. At the same time, it drives the adjustment of the minimum number of samples to increase, so that the clustering algorithm only accepts highly clustered data points as effective fault clusters. Through this global dynamic feedback mechanism, the algorithm exhibits strong false alarm suppression capability in harsh noise environments, while automatically smoothing regression under normal operating conditions and maintaining the original recognition sensitivity for early slight degradation.

[0062] It should be added that, in this embodiment, the initial preset neighborhood radius is set to 0.3, based on the following: after Z-score normalization, the multidimensional features follow a standard normal distribution, and the tolerance space of 0.3 times the standard deviation precisely covers the normal measurement uncertainty of the sensor system. At the same time, it strictly defines the baseline fluctuation boundary to prevent very early minor degradation shifts from being misjudged as normal. In this embodiment, the initial preset minimum sample number is set to 5, based on the integration of spatial and temporal dimensions: in the spatial dimension, it must be greater than the number of 3-dimensional features to resist high-dimensional random walk aggregation; in the temporal dimension, combined with a 5-minute sampling frequency, 5 consecutive points characterize a 25-minute continuous physical process, which matches the aforementioned 15 to 30-minute meteorological or state evolution cycle. Implementers can fine-tune the initial parameters according to the data distribution characteristics and typical electromagnetic background of the actual down conductor location, referring to the above criteria.

[0063] At this point, the adjusted minimum number of samples and neighborhood radius have been obtained.

[0064] Step S105: Cluster the real-time feature vector based on the adjusted parameters, and evaluate the anomaly probability of the current feature vector based on the clustering results, so as to realize real-time early warning of the grounding down conductor connection status.

[0065] Perform the following steps to achieve real-time analysis and early warning of connection status: Specifically, the feature vectors collected in real time are input into the DBSCAN model with adaptive parameters. Based on the adjusted neighborhood radius and minimum number of samples, the model determines the category of the current data point in the feature space in real time. This category can be clearly divided into core point clusters, representing stable and normal connection states; boundary point clusters, representing slight fluctuations or early degradation; and noise points, representing environmental interference or measurement anomalies.

[0066] Furthermore, the Mahalanobis distance between the current data point and the center of its core cluster is extracted as the spatial deviation of the current data point. The spatial deviation is normalized to the interval [0,1] using a negative exponential mapping function, and the result is used as the current anomaly probability.

[0067] The current anomaly probability is compared with the preset multi-level safety thresholds to trigger the corresponding closed-loop early warning command.

[0068] The preset multi-level safety thresholds include: attention threshold, alarm threshold and emergency threshold, which are set to 0.4, 0.7 and 0.9 respectively in this embodiment. The rationale is as follows: Due to the alternating temperature and humidity of shallow soil or the natural thermal noise of sensors, the normal measurement baseline will experience slight drift. The residual fluctuation probability after mapping is usually distributed between 0.1 and 0.3. Taking 0.4 as the attention threshold can effectively filter out false degradation caused by weather and can also sensitively trigger attention-level warnings during the incubation period of microscopic electrochemical corrosion on the metal contact surface of the down conductor. From a spatial statistical perspective, exceeding 0.7 means that the current vector has deviated from the normal cluster center by more than approximately three standard deviations of the confidence boundary. Physically, this corresponds to the down conductor connecting bolts being loosened due to stress, or the contact surface oxidation causing a substantial change in impedance. When the probability is extremely close to 1, it indicates that the multidimensional feature has entered the extreme high-voltage zone of historical severe fault samples, representing that the down conductor has undergone substantial fracture due to long-term corrosion or external damage, and the lightning protection discharge channel is severely damaged. Based on this, the system sets 0.9 to trigger an emergency-level warning and recommends immediate repair. Implementers can flexibly adjust the threshold according to the lightning protection safety importance requirements of different voltage level lines and the local historical thunderstorm frequency.

[0069] This completes the real-time analysis, monitoring, and early warning of the grounding down conductor connection status, ensuring that maintenance personnel can promptly grasp the health status of the grounding down conductor and guarantee the lightning protection and operational safety of the transmission line.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for real-time analysis and monitoring of the connection status of a grounding down conductor, characterized in that, include: Real-time acquisition of standardized multi-dimensional data of grounding down conductors, and construction of real-time feature vectors; The current window is divided based on the current time. The local anomaly factor algorithm is used to detect outliers in each dimension of the data within the current window, identifying outliers in each dimension. The timestamps corresponding to the outliers are extracted and sorted in ascending order to obtain an ordered sequence of outlier timestamps for each dimension. If the number of outliers in a certain dimension within the current window is less than 2, the temporal pulse feature value for that dimension is directly set to 0. If the number of outliers in a certain dimension within the current window is greater than or equal to 2, the preceding and following time-series data segments for each outlier's corresponding timestamp are extracted. The first mean of the monitored values ​​in the preceding time-series data segment and the second mean of the monitored values ​​in the following time-series data segment are calculated. The absolute difference between the first and second means is taken as the amplitude deviation of that outlier. The maximum amplitude deviation of all outliers in that dimension within the current window is extracted as the maximum amplitude deviation. The average value of all adjacent timestamp intervals in the outlier timestamp sequence for each dimension is calculated, and the average value is normalized to obtain the first feature. The maximum amplitude deviation is normalized, and the normalized maximum amplitude deviation is subtracted from the constant 1 to obtain the second feature; the first feature is multiplied by the second feature to obtain the time-series pulse feature values ​​of each dimension. Select a baseline dimension from the multidimensional data and use the remaining dimensions as correlation dimensions; perform a translational comparison between the outlier timestamp sequences of the baseline dimension and any correlation dimension to obtain the number of synchronous outliers, and calculate the outlier coupling correction factor by combining the difference in the temporal pulse eigenvalues ​​of the two. The time-series pulse feature values ​​of the baseline dimension are weighted based on the minimum value of the outlier coupling correction factor to obtain the noise confidence, and noise saliency windows are distinguished based on the empirical classification threshold. The global noise interference intensity is obtained by statistically analyzing the proportion of significant noise windows within a preset historical time period. The initial preset neighborhood radius and initial preset minimum sample number of the spatial clustering model are dynamically adjusted by utilizing the global noise interference intensity. The spatial clustering model with adjusted parameters is used to cluster real-time feature vectors. Based on the clustering results, the abnormal probability of the current feature vectors is evaluated to achieve real-time early warning of the grounding down conductor connection status.

2. The method for real-time analysis and monitoring of the connection status of a grounding down conductor according to claim 1, characterized in that, The step of shifting and comparing the outlier timestamp sequences of the baseline dimension with any associated dimension to obtain the number of synchronized outliers includes: While keeping the element index boundaries of the outlier timestamp sequence in the baseline dimension unchanged, the outlier timestamp sequence in any associated dimension is traversed and shifted in the direction of increasing or decreasing index by a set shift step size. Under each shift amount, valid point pairs within the index overlap interval of the outlier timestamp sequence in the baseline dimension and any associated dimension after shift are extracted, and the average absolute difference of the timestamps of all valid point pairs is calculated as the time difference. The shift amount corresponding to the minimum time difference is determined as the optimal shift amount, and the total number of valid point pairs under the optimal shift amount is extracted and denoted as the number of synchronized outliers.

3. The method for real-time analysis and monitoring of the connection status of a grounding down conductor according to claim 1, characterized in that, The calculation of the outlier coupling correction factor includes: Calculate the ratio of the number of synchronized outliers to the total number of outliers in the baseline dimension; calculate the absolute value of the difference between the temporal pulse eigenvalue of the baseline dimension and the temporal pulse eigenvalue of any associated dimension, and perform an exponential operation with the natural constant as the base and the opposite of the absolute value as the exponent to obtain the feature difference attenuation term; multiply the ratio by the feature difference attenuation term to obtain the outlier coupling correction factor between the baseline dimension and any associated dimension.

4. The method for real-time analysis and monitoring of the connection status of a grounding down conductor according to claim 1, characterized in that, The process of obtaining noise confidence and distinguishing noise salient windows based on empirical classification thresholds includes: Extract the minimum outlier coupling correction factor among the baseline dimension and all related dimensions within the current window, and denote it as the comprehensive coupling correction factor; calculate the noise confidence level of the current window. , In the formula, Represents the temporal pulse characteristic value of the baseline dimension within the current window; Indicates the comprehensive coupling correction factor; when When the noise level exceeds the empirical classification threshold, the current window is marked as a noisy window.

5. The method for real-time analysis and monitoring of the connection status of a grounding down conductor according to claim 1, characterized in that, The method of dynamically adjusting the initial preset neighborhood radius and initial preset minimum sample number of the spatial clustering model using global noise interference intensity includes: Calculate the adjusted minimum sample size parameter , In the formula, The initial minimum number of samples is preset; Given the global noise interference intensity; calculate the adjusted neighborhood radius. , In the formula, The initial preset neighborhood radius.

6. The method for real-time analysis and monitoring of the connection status of a grounding down conductor according to claim 1, characterized in that, The method of clustering real-time feature vectors using a spatial clustering model with adjusted parameters, and evaluating the anomaly probability of the current feature vectors based on the clustering results to achieve real-time early warning of the grounding down conductor connection status, includes: The spatial clustering model employs the density-based spatial clustering algorithm DBSCAN. After clustering the real-time feature vectors, the Mahalanobis distance between the current feature vector and the center of its corresponding core cluster is extracted as the spatial deviation of the current feature vector. The spatial deviation is then normalized to a negative exponential mapping function. Within the interval, the anomaly probability of the current feature vector is obtained; when the anomaly probability exceeds the preset multi-level security threshold, the corresponding warning instruction is triggered.

7. The method for real-time analysis and monitoring of the connection status of a grounding down conductor according to claim 1, characterized in that, The method further includes: The multi-dimensional data specifically includes: grounding resistance, leakage current, and soil resistivity; wherein, the baseline dimension is grounding resistance; and the associated dimensions include leakage current and soil resistivity.

Images