Data transmission link anomaly detection method and system based on intelligent fusion terminal

Abstract

This invention relates to the field of electrical digital data processing, and more particularly to a method and system for detecting anomalies in data transmission links based on intelligent fusion terminals. The method includes: acquiring transformer and meter current data under normal operating conditions of the intelligent fusion terminal; preprocessing this data into a training sample sequence; extracting and correcting pseudo-limit samples to obtain a corrected limit sample sequence; adaptively classifying current levels through sample density analysis; determining the fluctuation amplitude of the current difference at each level; calculating a dynamic threshold by fusing the line loss rate with this amplitude; constructing an anomaly detection model and embedding it into the terminal; performing anomaly detection on real-time current time-series data before data cleaning; and outputting link distortion results. This invention, through adaptively classifying the reference current levels and constructing a hierarchical anomaly detection model, eliminates latent anomalies caused by synchronous distortion in dual links, effectively adapting to the operating conditions of dynamically changing reference currents.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing. In particular, it relates to a method and system for detecting anomalies in data transmission links based on intelligent converged terminals. Background Technology

[0002] With the rapid development of smart grids, intelligent converged terminals have become core equipment for power grid data acquisition, processing, and transmission. Their operational status directly affects the accuracy of power grid monitoring and the timeliness of fault early warning. Intelligent converged terminals bear the important responsibility of integrating data from multiple sources and supporting subsequent core business operations such as line loss rate calculation, and are a fundamental support for ensuring the stable and efficient operation of the power grid.

[0003] As the core channel for intelligent converged terminals to acquire valid data, the stability of the data transmission link directly determines the data quality. Once the link malfunctions and causes data distortion, it will not only interfere with the accuracy of subsequent business processes such as line loss rate calculation and data fusion, but may also lead to misjudgment of the power grid's operating status, thereby delaying fault warning or triggering unnecessary emergency responses, bringing additional costs to power grid operation and maintenance, and even affecting power supply reliability.

[0004] When transformers and meters transmit current data to smart converged terminals, on the one hand, due to the dynamic changes in the reference current, the fluctuation base of the current difference in the data transmitted between the transformer and the smart meter under different reference currents has an inherent difference, making a single fixed threshold unsuitable for diverse operating conditions; on the other hand, data distortion may occur in both the transformer and meter-to-terminal dual transmission links. It is difficult to accurately distinguish complex situations such as data transmission link distortion, abnormal line loss rate, and dual-link distortion by relying solely on single data comparison or existing anomaly detection methods. Ultimately, this results in insufficient reliability and accuracy in data transmission link anomaly detection, making it impossible to timely and accurately identify single-link or dual-link data distortion problems, and failing to meet the needs of refined operation and maintenance of smart grids. Summary of the Invention

[0005] To address the issues of poor adaptability of a single fixed threshold due to dynamic changes in the reference current when transformers and meters transmit current data to a smart fusion terminal, and the potential for simultaneous distortion of both links, existing judgment methods struggle to distinguish complex abnormal situations, resulting in insufficient reliability and accuracy in link anomaly detection, this invention provides solutions in the following aspects.

[0006] In the first aspect, the data transmission link anomaly detection method based on the intelligent fusion terminal includes: acquiring transformer current data and meter current data under normal operating conditions of the intelligent fusion terminal, preprocessing them, and using them as training sample sequences; extracting extreme sample sequences from the training sample sequences and correcting pseudo extreme samples that violate the rules of change; based on the corrected extreme sample sequences, identifying abrupt change density through sample density analysis, adaptively classifying current levels according to the abrupt change density, and selecting the maximum value of the maximum current difference component in the extreme sample set corresponding to each current level as the current difference fluctuation amplitude; calculating the line loss rate based on historical current data without anomalies, calculating a dynamic threshold by fusing the line loss rate and the current difference fluctuation amplitude corresponding to each current level, constructing a data distortion anomaly detection model containing the dynamic threshold, embedding the data distortion anomaly detection model into the intelligent fusion terminal, and performing anomaly detection on the real-time received transformer and meter current time series data before the terminal executes the data cleaning process, and outputting the detection result of whether data distortion has occurred in the data transmission link.

[0007] Preferably, the step of obtaining the training sample sequence includes: The absolute difference between transformer current data and meter current data at the same time is calculated to obtain the current difference time series. Based on the transformer current data as the reference current, a two-dimensional sample vector is constructed by combining the current difference data of the corresponding time series. After sorting in ascending order according to the reference current, a training sample sequence is formed.

[0008] Preferably, the step of obtaining the corrected limit sample sequence includes: Extract the maximum current difference component corresponding to each reference current from the training sample sequence, construct the limit sample vector and arrange it in ascending order of reference current to form the limit sample sequence; The limit sample sequence is preprocessed by iterating through the maximum current difference components of adjacent limit samples. If the maximum current difference component of the previous limit sample is greater than the maximum current difference component of the current limit sample, the maximum current difference component of the current limit sample is corrected to the maximum current difference component of the previous limit sample, thus obtaining the corrected limit sample sequence.

[0009] Preferably, the current level is adaptively divided according to the mutation density, including: The neighborhood sample range of each limit sample in the preprocessed limit sample sequence is preset. The current difference distance and reference distance between each limit sample and its neighborhood sample are calculated. The mean current difference distance and the mean reference distance are combined with the exponential function and the bandwidth coefficient to calculate the sample density of each limit sample, thus forming a sample density sequence. The sample density sequence is first-order differencing to obtain a difference sequence. The elbow method is used to identify the inflection point of the difference sequence and determine the mutation density. The reference current corresponding to the limit sample where the first-order difference is greater than the mutation density is used as the dividing point. The reference current is divided into continuous intervals to obtain several reference current levels.

[0010] Preferably, the sample density is calculated using the following methods: Taking any extreme sample as the sample to be analyzed, calculate the absolute value of the straight-line distance between the sample to be analyzed and each neighboring sample on the vertical axis to obtain the current difference distance. Calculate the average value of all current difference distances corresponding to the sample to be analyzed to obtain the mean value of the current difference distance. Calculate the absolute value of the straight-line distance between the sample to be analyzed and each neighboring sample on the horizontal axis to obtain the baseline distance between the sample to be analyzed and each neighboring sample. Calculate the average value of all baseline distances corresponding to the sample to be analyzed to obtain the mean baseline distance. The ratio of the mean current difference distance to the mean reference distance is used as the relative degree of the current difference fluctuation amplitude of the sample to be analyzed; the relative degrees of all extreme samples are integrated to form a relative degree sequence, and the maximum value of the relative degree sequence is extracted as the bandwidth coefficient; Multiply the relative degree by the reciprocal of the bandwidth coefficient and take the negative value. Substitute the result into the exponential function to calculate the sample density of the sample to be analyzed.

[0011] Preferably, the dynamic threshold is calculated using the following method: The line loss rate is taken as the ratio of the baseline line loss rate to the maximum line loss rate in the historical current data. Multiplying this ratio by a preset optimization weighting coefficient yields the line loss rate correction coefficient. The maximum value between the line loss rate correction coefficient and 1 is selected and multiplied by the current difference fluctuation amplitude under the current baseline current level to obtain the dynamic threshold.

[0012] Preferably, the construction of the data distortion anomaly detection model including dynamic thresholds includes: Set up hierarchical judgment steps: First-level judgment: Traverse all timestamps of real-time current data pairs within the current data window. , For the transformer output current, For electricity meters to measure current, the response satisfies the following at any timestamp. If so, it is directly judged as a distortion anomaly; Second-level judgment: If the first-level judgment is not triggered, the standard deviation of the current difference at each timestamp and the standard deviation of all current differences within the window are calculated as the real-time current difference standard deviation. If the real-time current difference standard deviation is greater than the dynamic threshold, it is judged as a type of distortion anomaly. Third-level judgment: In response to the failure of the second-level judgment, calculate the current line loss rate of the current window, and then calculate the relative deviation between the current line loss rate and the baseline line loss rate. In response to the relative deviation being greater than the preset abnormal fluctuation threshold, it is judged as a type II distortion anomaly. If none of the above three levels of judgment are triggered, it is determined that there is no distortion or abnormality in the data transmission link, and the real-time current data in the current window is valid.

[0013] Preferred options also include: Based on the detection results, historical current data are verified. The misjudgment rates of Type I and Type II distortion anomalies generated during the verification process are statistically analyzed. These rates correspond to the judgment deviations under different anomaly judgment logics in the model. The statistically obtained misjudgment rates are then used as feedback to optimize the relevant parameters in the model.

[0014] Secondly, a data transmission link anomaly detection system based on a smart converged terminal includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned data transmission link anomaly detection method based on a smart converged terminal is implemented.

[0015] The present invention has the following effects: 1. This invention adaptively classifies the reference current level by analyzing the sample density based on the extreme sample sequence, and calculates the dynamic threshold by integrating the line loss rate and the fluctuation amplitude of the current difference at each level. This replaces the traditional single fixed threshold, effectively adapting to the dynamic changes in the reference current and significantly improving the adaptability of anomaly detection to complex power grid operating conditions.

[0016] 2. This invention, through the construction of a hierarchical anomaly detection model, adds a third-level judgment based on line loss rate verification on the basis of current difference fluctuation analysis. It identifies hidden anomaly scenarios such as dual-link synchronization distortion by relying on the relative deviation index of line loss rate. This breaks through the technical limitation of existing judgment methods that are difficult to distinguish complex anomalies, and significantly improves the accuracy and reliability of data transmission link anomaly detection. Attached Figure Description

[0017] Figure 1 This is a flowchart of steps S1-S3 in the data transmission link anomaly detection method based on a smart fusion terminal according to an embodiment of the present invention.

[0018] Figure 2 This is a structural block diagram of the data transmission link anomaly detection system based on an intelligent fusion terminal according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0020] Reference Figure 1 The data transmission link anomaly detection method based on intelligent converged terminals includes steps S1-S3, as follows: S1: Obtain transformer current data and meter current data under normal operating conditions of the intelligent fusion terminal, and after preprocessing, use them as training sample sequences. Extract the limit sample sequences from the training sample sequences and correct the pseudo limit samples that violate the rules.

[0021] Obtain the current value output by the transformer of the intelligent converged terminal. The current value fed back by the smart meter A transformer time-series current sequence is constructed based on the current output of the transformer, and a meter time-series current sequence is constructed based on the current feedback from the smart meter. The transformer time-series current sequence is then... The current value is at the sampling time. The current value of the transformer collected The first of the time-series current sequence of the electricity meter The current value is at the sampling time. The current value collected from the smart meter .

[0022] To ensure the validity of subsequent model training data, the output current value of the transformer needs to be... The current value fed back by the smart meter A triple validity verification process is implemented to filter valid data without link distortion anomalies: The first step is physical link verification, which confirms the normal physical status of the transmission link between the transformer, smart meter, and smart converged terminal through manual inspection and meter reading records; after passing the physical link verification, the second step of verification is performed. If the verification fails, the current data is directly removed and not included in subsequent training samples; the second step is line loss rate verification, which ensures that the line loss rate corresponding to the data is within a preset stable range; the third step is equipment status verification, which reads the equipment status codes reported by the transformer acquisition equipment and smart meters based on the equipment status codes of the transformer and smart meters. When the status code is an abnormal working state such as fault identifier, communication abnormal identifier, sampling abnormal identifier, alarm identifier, etc., the data at that moment is determined to be fault alarm abnormal data and is directly removed and not included in the training samples; the equipment status verification is only passed when the equipment status code is in normal working state; abnormal data with fault alarms are excluded.

[0023] After validity verification, the absolute difference between the transformer current value and the meter current value was calculated using the same timestamp for the two types of current time series data. This absolute difference time series data was defined as the current difference. ,in Indicates the first The current difference at each time point is used to obtain the current difference time series.

[0024] Using the transformer current as the reference current, the first A two-dimensional sample vector is constructed from the reference current value and the current difference at each time step. Each two-dimensional sample vector corresponds to a valid training sample; then, according to the baseline current... The numerical values ​​of all training samples are sorted in ascending order to form an ordered sequence of training samples. This provides a standardized data foundation for subsequent benchmark current level classification and model training.

[0025] It should be noted that the core purpose of this step in classifying the reference current level is to address the technical problem of insufficient statistical validity of current difference samples corresponding to a single current value: during power grid operation, the reference current output by the transformer... As a continuously changing physical quantity, if the normal fluctuation range of the current difference is calibrated separately for a single reference current value, there are two problems. First, it is difficult to control the sample size of the current difference corresponding to a single current value, making it impossible to accurately determine its normal fluctuation range through statistical methods. Second, it is easily affected by isolated noise such as pulse interference, causing the fluctuation range calibration to be distorted. However, within a continuous reference current range, the fluctuation pattern of the current difference has statistical consistency. By classifying the levels, the samples within this range can be aggregated to ensure that a sufficient sample size is obtained to guarantee the statistical validity of the fluctuation range calculation, while also diluting the interference of isolated noise on the statistical results.

[0026] Due to noise such as pulse interference that may exist during the power grid data acquisition process, the maximum current difference component of some samples in the limit sample sequence violates the physical law that the maximum current difference component synchronously ascends when the reference current increases, forming pseudo limit samples. If this is not corrected, it will affect the accuracy of the subsequent reference current level classification.

[0027] Specifically, the maximum current difference component is extracted from the reference current value for each training sample sequence, and a vector is constructed. As the limiting samples, the limiting samples are arranged in ascending order according to the reference current to obtain the limiting sample sequence.

[0028] The limit sample sequence is traversed sequentially, and the maximum current difference component in adjacent limit samples is compared. If the maximum current difference component of the previous limit sample is greater than the maximum current difference component of the current limit sample, the current limit sample is determined to be a pseudo-limit sample, and the maximum current difference component in the pseudo-limit sample vector is corrected to the maximum current difference component of the previous limit sample, thus completing the data cleaning. After the traversal is completed, a preprocessed limit sample sequence is obtained. This sequence ensures the regularity of the maximum current difference component changing with the ascending order of the reference current, providing reliable data support for subsequent sample density calculation and accurate classification of the reference current level.

[0029] S2: Based on the corrected limit sample sequence, the mutation density is identified through sample density analysis. The current level is adaptively divided according to the mutation density, and the maximum value of the maximum current difference component in the limit sample set corresponding to each current level is selected as the current difference fluctuation amplitude.

[0030] Each extreme sample in the preprocessed extreme sample sequence As the data foundation, the reference current component of the sample vector Using the maximum current difference component as the x-axis Establish a coordinate system using the vertical axis as the ordinate.

[0031] For the preprocessed limit sample sequence, a neighborhood sample range is first preset for each limit sample: taking any limit sample as the sample to be analyzed, and using the reference current of the sample to be analyzed as the center, a preset fixed neighborhood distance is defined before and after it. All limit samples within this range are selected as neighborhood samples of the corresponding limit sample, and it is ensured that the number of neighborhood samples of each limit sample is not less than 10, so as to ensure the statistical validity of subsequent calculations.

[0032] Based on the range of neighboring samples, the current difference distance and reference distance between each limiting sample and the neighboring sample are calculated respectively: the absolute value of the straight-line distance between the sample to be analyzed and the neighboring sample on the maximum current difference component is calculated to obtain the current difference distance; the average value of all current difference distances corresponding to the sample to be analyzed is calculated to obtain the mean current difference distance; the absolute value of the straight-line distance between the sample to be analyzed and the neighboring sample on the reference current component is calculated to obtain the reference distance; the average value of all reference distances corresponding to the sample to be analyzed is calculated to obtain the mean reference distance. The ratio of the mean current difference distance to the mean reference distance is used as the relative degree of the current difference fluctuation amplitude of the sample to be analyzed. The relative degrees of all extreme samples are integrated to form a relative degree sequence, and the maximum value of the relative degree sequence is extracted as the bandwidth coefficient. The relative degree is multiplied by the reciprocal of the bandwidth coefficient and the negative value is taken. The result is substituted into the exponential function for calculation to obtain the sample density of the sample to be analyzed.

[0033] Specifically, the sample density satisfies the following relationship: ; In the formula, This represents the sample density of the sample to be analyzed, and its numerical range is... , This represents the bandwidth coefficient, and its function is to avoid [the following text is incomplete and likely refers to a different topic]: A value that is too large results in a low degree of differentiation in density values. Let represent the mean distance between the current difference of any limiting sample and all its neighboring samples. Let represent the mean of the baseline distances between any limiting sample and all its neighboring samples. Indicated by An exponential function with base 0.

[0034] It should be noted that the bandwidth coefficient is a key adjustment parameter in the calculation of sample density. Its core function is to avoid the ratio of the mean current difference distance to the mean reference distance being too large, which would result in the density value having too small a distinction, and to ensure that the sample density can accurately represent the abrupt boundary of the reference current level.

[0035] The mean current difference distance is used to characterize the consistency of fluctuations between any extreme sample and its neighboring samples in the dimension of the maximum current difference component. The smaller the value, the more consistent the current difference fluctuations of the samples in this local range are, and the greater the probability that the corresponding samples belong to the same reference current level.

[0036] The core function of the mean of the reference distance is to reduce the interference of uneven distribution of limiting samples in the reference current direction on sample density calculation. Since it is difficult to achieve an ideally uniform and equally spaced distribution of limiting samples in the reference current direction, and the total number of neighboring samples within a set distance range varies for different limiting samples, the mean distance can effectively reduce the adverse effects of such distribution differences on the density calculation results. For example, when a limiting sample is not at the switching boundary of the reference current level, but its distance from neighboring samples is large due to sparse sample distribution, the mean distance can be reduced. It will show a large value, while the corresponding value for this sample is... It will also increase synchronously, at which point... The ratio of these ratios in the sample density calculation can significantly enhance the reliability and reference value of the sample density.

[0037] The sample density of each extreme sample is calculated to form a sample density sequence. The first-order difference of the sample density sequence is performed to obtain the difference sequence. The elbow method is used to identify the inflection point of the difference sequence and determine the mutation density. The reference current corresponding to the extreme sample with the first-order difference greater than the mutation density is used as the dividing point. The reference current is divided into continuous intervals to obtain several reference current levels. Each level corresponds to a continuous reference current interval with a stable current difference fluctuation law. This provides an accurate basis for the calibration of the current difference fluctuation amplitude of each level and the design of dynamic thresholds.

[0038] It should be noted that the elbow technique is a well-known technique in the field and will not be described in detail.

[0039] S3: Calculate the line loss rate based on historical current data without abnormalities, calculate the dynamic threshold by integrating the line loss rate with the current difference fluctuation amplitude corresponding to each current level, construct a data distortion anomaly detection model containing the dynamic threshold, embed the data distortion anomaly detection model into the intelligent fusion terminal, and perform anomaly detection on the real-time received current time series data of transformers and meters before the terminal executes the data cleaning process, and output the detection result of whether data distortion has occurred in the data transmission link.

[0040] When implementing the technical solution of this application, the basic calculation method for the line loss rate can adopt the technical solution disclosed in the applicant's prior application (application number CN202510880120.X, invention title: A Smart Terminal Supporting Multi-Parameter Fusion Sensing), specifically the implementation method of "rolling calculation of line loss rate within a set period" described in paragraph 53 of the prior application specification. This part contains common technical features of this application and prior applications, and its specific implementation details have been fully disclosed in the prior applications. Therefore, this application will not repeat the detailed description of this part.

[0041] Based on the aforementioned common technical features, this application further improves the selection of the baseline loss rate and the calculation of the dynamic threshold. The specific steps are as follows: The line loss rate calculated using the previous window of no abnormalities is used as the baseline line loss rate. The ratio of the baseline line loss rate to the maximum line loss rate in the historical current data is calculated. The ratio is multiplied by the preset optimization weight coefficient to obtain the line loss rate correction coefficient. The maximum value between the line loss rate correction coefficient and the value of 1 is selected. The maximum value is then compared with the current difference fluctuation amplitude under the current baseline current level to obtain the dynamic threshold.

[0042] Specifically, the dynamic threshold satisfies the following relationship: ; In the formula, Indicates a dynamic threshold. Represents the maximum value function. Indicates the baseline loss rate; This represents the maximum line loss rate in historical current data. This represents the preset optimization weight coefficients, with initial empirical values ​​set to [value missing]. , This indicates the amplitude of the current difference fluctuation at the current reference current level.

[0043] The optimization of the weighting coefficients is based on historical current data from routine power grid operation under non-extreme peak-valley loads and without significant topological changes. This data reveals that the ratio of the baseline line loss rate to the historical maximum line loss rate typically stabilizes between 0.6 and 0.8 under normal operating conditions. Based on this experience, the initial empirical value is set as follows: At that time, the corresponding line loss rate correction coefficient is in the range of 0.72 to 0.96. Secondly, the initial empirical value, verified through multiple sets of routine operating condition simulations, can control the initial misjudgment rate of a type of anomaly within a preset empirical threshold of 3%, thus avoiding the risk of errors caused by [other factors]. Too small an initial empirical value can lead to an overly strict threshold and a high false positive rate. Conversely, too large an initial value can lead to an overly broad threshold and an increased risk of missed judgments.

[0044] Line loss rate is a key indicator characterizing the consistency between power supply and sales. Its value is directly related to the operating conditions of the power grid and the status of the data transmission link. The higher the current line loss rate, the more significant the power loss characteristics or data fluctuation characteristics of the power grid itself under normal operating conditions. At this time, the normal fluctuation range of the real-time current difference should be expanded accordingly. Based on this physical characteristic, the current difference fluctuation amplitude of the corresponding reference current level is adjusted by the line loss rate correction coefficient during the dynamic threshold calculation process, so that the dynamic threshold increases synchronously, thereby improving the tolerance for real-time current difference data and avoiding misjudging the normal current difference fluctuation caused by the increase in line loss rate as abnormal data transmission link distortion, thus ensuring the accuracy of anomaly detection and the adaptability of operating conditions.

[0045] Constructing a data distortion anomaly detection model incorporating dynamic thresholds involves the following steps: First-level judgment: Traverse all real-time current data pairs corresponding to all timestamps within the current data window. ,in This is the transformer current value (supply current). This represents the meter current value (electricity sales current). If any timestamp satisfies... If so, it is directly determined that the data transmission link has a distortion anomaly.

[0046] The physical basis for this level of determination is: In the actual operation of the power grid, energy loss due to line losses objectively exists. Under normal operating conditions, the supply current output by the transformer must be greater than the electricity sales current measured by the meter. This is a fundamental prerequisite for operating without abnormal conditions. When If the above physical laws are violated, it can be directly identified as a data transmission link abnormality causing data distortion.

[0047] If the first-level judgment is not triggered, that is, all timestamps within the window meet the condition. First, perform the pre-parameter calculation: for each timestamp within the window, calculate the absolute value of the difference between the transformer current value and the meter current value to obtain the current difference. Then, combine all timestamps within the window and use the standard deviation of the current difference corresponding to each timestamp as the real-time current difference standard deviation. The real-time current difference standard deviation is used to quantify the drastic fluctuation of the current difference within the current window.

[0048] The second level of judgment: If the standard deviation of the real-time current difference exceeds the dynamic threshold, an anomaly is determined, and this type of anomaly is marked as a distortion anomaly. The core logic of this level of judgment is: the dynamic threshold integrates the baseline loss rate and the current difference fluctuation amplitude of the corresponding current level, representing the maximum normal fluctuation boundary of the current difference under the current operating conditions. When the standard deviation of the real-time current difference exceeds the maximum normal fluctuation boundary, it indicates that the current difference fluctuation has exceeded the reasonable range under normal operating conditions, and can be judged as distortion caused by an abnormal data transmission link.

[0049] If the second-level judgment is not triggered, i.e., the standard deviation of the real-time current difference is less than or equal to the dynamic threshold, then the line loss rate auxiliary verification stage is entered. It should be noted that the premise of this model is that the current value transmitted from the transformer to the terminal is not distorted. Therefore, even if the standard deviation of the real-time current difference is greater than the dynamic threshold, further verification through the line loss rate is still required to rule out hidden anomalies such as dual-link synchronization distortion.

[0050] The third level of judgment: First, calculate the line loss rate corresponding to the current window and define it as the current line loss rate; then calculate the relative deviation between the current line loss rate and the baseline line loss rate, i.e. ,in This indicates the current line loss rate. Indicates the baseline loss rate. This represents the threshold for abnormal fluctuations, with an initial empirical value set at 0.03. It is used in response to a relative deviation greater than... If the error occurs, it is determined that a data transmission link distortion anomaly has occurred, and this anomaly type is defined as a Type II distortion anomaly.

[0051] Abnormal fluctuation threshold The initial empirical value was set at 0.03, or 3%. At least six months of historical power grid operation data without anomalies were collected, covering scenarios such as normal load, mild peak-valley fluctuations, and normal equipment wear. The relative deviation of the line loss rate for each data window from the corresponding baseline line loss rate was calculated. It was found that under anomaly-free operating conditions, the line loss rate is affected by factors such as small load fluctuations, changes in ambient temperature, and minor equipment wear. Its relative deviation is generally stable within ±3%, and this range can characterize the normal natural fluctuation range of the line loss rate. Furthermore, through extensive simulation experiments, it was verified that when… When the equipment is at 3%, it can effectively distinguish between normal fluctuations in line loss rate and abnormal deviations in line loss rate caused by dual-link synchronization distortion, and can control the initial misjudgment rate of the second type of distortion anomaly within the 5% empirical threshold.

[0052] If none of the above-mentioned first-level, second-level, and third-level judgments are triggered, it is determined that no data distortion abnormality has occurred in the current data transmission link, and the real-time current data pair in the current window is valid data.

[0053] Anomaly detection is performed on historical current data and the corresponding anomaly detection results are output. The accuracy of the data distortion anomaly detection model is calculated based on these results, and then parameter optimization feedback is generated to achieve adaptive updating of the model parameters. The specific logic and implementation steps are as follows: The misclassification rates for Class I and Class II distortion anomalies are calculated separately to quantify the judgment bias under different judgment logics of the model, where: Taking a type of distortion anomaly as an example, specifically, a type of distortion anomaly satisfies the following relationship: ; In the formula, This represents the false positive rate for a type of distortion anomaly. The total number of judgments represents the number of times the input model performs anomaly judgments on historical current data. An anomaly judgment is recorded once for data within a window. The number of false positives represents the number of times the anomaly judgment result is incorrect. If the data distortion anomaly detection model determines that an anomaly has occurred and the anomaly type is a type of distortion anomaly, but the corresponding historical current data does not show any data distortion anomalies caused by the data transmission link, it is recorded as a false positive.

[0054] Similarly, the false positive rate for Type II distortion anomalies is calculated using the same logic as for Type I distortion anomalies. Since the false positive rate for Type I distortion anomalies is greater than the empirical value (set to 3% in this embodiment), the optimized weight coefficients are updated. The updated optimized weight coefficients satisfy the following relationship: ; In the formula, This represents the updated optimized weight coefficients. This represents the misjudgment rate of a type of distortion anomaly. This represents the maximum empirical value of the actually allowed optimizable weight coefficients, for example... ; This represents the optimized weighting coefficient.

[0055] Based on the physical laws of power grid operation, the statistical characteristics of historical current data, and the constraints of anomaly detection accuracy: under extreme power grid line loss conditions, it is necessary to ensure dynamic thresholds. The fluctuation range shall not exceed the maximum normal fluctuation range of the current difference for the corresponding reference current level. For example, it is typically the fluctuation amplitude of the current difference for that level. Two to three times higher, after analyzing at least six months of historical current data without abnormalities under all operating conditions, it was found that when Even at that time Approaching 1, the line loss rate correction factor The maximum is approximately 3, combined with After calculation, Maximum not exceeding It will not be because Excessive size can lead to missed anomalies, and can also avoid [further issues]. Unlimited thresholds lead to uncontrolled thresholds; however, this value can stably control the false positive rate of a certain type of distortion anomaly within a preset 1%, and the false negative rate. It adapts to complex operating conditions such as peak and valley loads and seasonal changes in the power grid, and takes into account both the accuracy of anomaly detection and the adaptability to operating conditions. It is the optimal empirical upper limit value determined after multiple rounds of iterative calibration.

[0056] It should be noted that the high misjudgment rate of Type I distortion anomalies indicates that the current dynamic threshold is too strict in judging the anomalies of current difference fluctuations. It is easy to misjudge current difference fluctuations under normal operating conditions as Type I distortion anomalies. Therefore, the judgment standard of this anomaly needs to be relaxed, that is, the value of the dynamic threshold needs to be increased to adapt to the normal fluctuation range of current difference.

[0057] In response to the fact that the false positive rate of Type II distortion anomalies is greater than the empirical value, in this embodiment, it is set to 5%, and the anomaly fluctuation coefficient is updated. The updated optimized weight coefficient satisfies the following relationship: ; In the formula, This represents the updated optimized weight coefficients. This represents the misclassification rate of type II distortion anomalies. This represents the optimized weighting coefficient.

[0058] It should be noted that the high misjudgment rate of Type II distortion anomalies indicates that the current anomaly fluctuation threshold sets an overly strict anomaly judgment standard for the deviation between the current line loss rate and the baseline line loss rate. This makes it easy to misjudge normal fluctuations in the line loss rate as Type II distortion anomalies. Therefore, the anomaly judgment standard needs to be relaxed, i.e., the value of the anomaly fluctuation threshold needs to be increased to match the normal fluctuation range of the line loss rate and reduce the probability of misjudgment.

[0059] Finally, the data distortion anomaly detection model is embedded into the existing patent. After self-optimization of the model, accurate detection of anomalies in the data transmission link from transformers and meters to smart fusion terminals is achieved.

[0060] This invention also provides a data transmission link anomaly detection system based on an intelligent converged terminal. For example... Figure 2As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the data transmission link anomaly detection method based on an intelligent fusion terminal according to the first aspect of the present invention. The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, the settings and functions of which are known in the art and therefore will not be described further here.

[0061] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting anomalies in data transmission links based on intelligent converged terminals, characterized in that, include: The transformer current data and meter current data of the intelligent fusion terminal under normal working conditions are obtained and preprocessed as training sample sequences. The extreme sample sequences in the training sample sequences are extracted and the pseudo extreme samples that violate the rules of change are corrected. Based on the corrected limit sample sequence, the mutation density is identified through sample density analysis. The current level is adaptively divided according to the mutation density, and the maximum value of the maximum current difference component in the limit sample set corresponding to each current level is selected as the current difference fluctuation amplitude. The line loss rate is calculated based on historical current data without anomalies. The dynamic threshold is calculated by combining the line loss rate with the current difference fluctuation amplitude corresponding to each current level. A data distortion anomaly detection model containing the dynamic threshold is constructed. The data distortion anomaly detection model is embedded into the intelligent fusion terminal. Before the terminal performs the data cleaning process, anomaly detection is performed on the real-time received current time series data of transformers and meters, and the detection result of whether data distortion has occurred in the data transmission link is output.

2. The data transmission link anomaly detection method based on intelligent converged terminal according to claim 1, characterized in that, The steps for obtaining the training sample sequence include: The absolute difference between transformer current data and meter current data at the same time is calculated to obtain the current difference time series. Based on the transformer current data as the reference current, a two-dimensional sample vector is constructed by combining the current difference data of the corresponding time series. After sorting in ascending order according to the reference current, a training sample sequence is formed.

3. The data transmission link anomaly detection method based on intelligent converged terminals according to claim 1, characterized in that, The steps for obtaining the corrected limit sample sequence include: Extract the maximum current difference component corresponding to each reference current from the training sample sequence, construct the limit sample vector and arrange it in ascending order of reference current to form the limit sample sequence; The limit sample sequence is preprocessed by iterating through the maximum current difference components of adjacent limit samples. If the maximum current difference component of the previous limit sample is greater than the maximum current difference component of the current limit sample, the maximum current difference component of the current limit sample is corrected to the maximum current difference component of the previous limit sample, thus obtaining the corrected limit sample sequence.

4. The data transmission link anomaly detection method based on intelligent converged terminal according to claim 1, characterized in that, Current levels are adaptively classified based on mutation density, including: The neighborhood sample range of each limit sample in the preprocessed limit sample sequence is preset. The current difference distance and reference distance between each limit sample and its neighborhood sample are calculated. The mean current difference distance and the mean reference distance are combined with the exponential function and the bandwidth coefficient to calculate the sample density of each limit sample, thus forming a sample density sequence. The sample density sequence is first-order differencing to obtain a difference sequence. The elbow method is used to identify the inflection point of the difference sequence and determine the mutation density. The reference current corresponding to the limit sample where the first-order difference is greater than the mutation density is used as the dividing point. The reference current is divided into continuous intervals to obtain several reference current levels.

5. The data transmission link anomaly detection method based on intelligent converged terminal according to claim 4, characterized in that, The calculation method for the sample density includes: Taking any extreme sample as the sample to be analyzed, calculate the absolute value of the straight-line distance between the sample to be analyzed and each neighboring sample on the vertical axis to obtain the current difference distance. Calculate the average value of all current difference distances corresponding to the sample to be analyzed to obtain the mean value of the current difference distance. Calculate the absolute value of the straight-line distance between the sample to be analyzed and each neighboring sample on the horizontal axis to obtain the baseline distance between the sample to be analyzed and each neighboring sample. Calculate the average value of all baseline distances corresponding to the sample to be analyzed to obtain the mean baseline distance. The ratio of the mean current difference distance to the mean reference distance is used as the relative degree of the current difference fluctuation amplitude of the sample to be analyzed; the relative degrees of all extreme samples are integrated to form a relative degree sequence, and the maximum value of the relative degree sequence is extracted as the bandwidth coefficient; Multiply the relative degree by the reciprocal of the bandwidth coefficient and take the negative value. Substitute the result into the exponential function to calculate the sample density of the sample to be analyzed.

6. The data transmission link anomaly detection method based on intelligent converged terminal according to claim 1, characterized in that, The calculation method for the dynamic threshold includes: The line loss rate is taken as the ratio of the baseline line loss rate to the maximum line loss rate in the historical current data. Multiplying this ratio by a preset optimization weighting coefficient yields the line loss rate correction coefficient. The maximum value between the line loss rate correction coefficient and 1 is selected and multiplied by the current difference fluctuation amplitude under the current baseline current level to obtain the dynamic threshold.

7. The data transmission link anomaly detection method based on intelligent converged terminal according to claim 1, characterized in that, The construction of the data distortion anomaly detection model including dynamic thresholds includes: Set up hierarchical judgment steps: First-level judgment: Traverse all timestamps of real-time current data pairs within the current data window. , For the transformer output current, For electricity meters to measure current, the response satisfies the following at any timestamp. If so, it is directly judged as a distortion anomaly; Second-level judgment: If the first-level judgment is not triggered, the standard deviation of the current difference at each timestamp and the standard deviation of all current differences within the window are calculated as the real-time current difference standard deviation. If the real-time current difference standard deviation is greater than the dynamic threshold, it is judged as a type of distortion anomaly. Third-level judgment: In response to the failure of the second-level judgment, calculate the current line loss rate of the current window, and then calculate the relative deviation between the current line loss rate and the baseline line loss rate. In response to the relative deviation being greater than the preset abnormal fluctuation threshold, it is judged as a type II distortion anomaly. If none of the above three levels of judgment are triggered, it is determined that there is no distortion or abnormality in the data transmission link, and the real-time current data in the current window is valid.

8. The data transmission link anomaly detection method based on intelligent converged terminal according to claim 1, characterized in that, Also includes: Based on the detection results, historical current data are verified. The misjudgment rates of Type I and Type II distortion anomalies generated during the verification process are statistically analyzed. These rates correspond to the judgment deviations under different anomaly judgment logics in the model. The statistically obtained misjudgment rates are then used as feedback to optimize the relevant parameters in the model.

9. A data transmission link anomaly detection system based on an intelligent converged terminal, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the data transmission link anomaly detection method based on a smart fusion terminal according to any one of claims 1-8.

Images