A power distribution cabinet load data abnormal backhaul checking method and system

Abstract

The application relates to a power distribution cabinet load data abnormal back transmission checking method and system, which comprises the following steps: obtaining local and remote load data, extracting a timestamp to establish an alignment mapping table, linearly interpolating a synchronous data pair for unmatched points; point-by-point comparison is performed to calculate a deviation difference sequence, the maximum deviation and the average deviation are extracted as historical baselines according to time period division; a difference threshold is determined according to the historical baseline, an abnormal score is obtained by using an isolated forest, and a distortion point is marked if the abnormal score exceeds a preset threshold and the deviation value exceeds the difference threshold by a preset multiple; transmission time delay within a preset time length before and after the distortion point is extracted, and a fluctuation amplitude is calculated, and if the fluctuation amplitude exceeds a preset fluctuation threshold and the peak time delay exceeds a preset peak threshold, the packet loss is determined and an abnormal period is recorded; a retransmission request is generated to receive and resend data and update the difference sequence; the classifier classifies and counts the packet loss abnormal proportion, and the packet loss is confirmed if the proportion exceeds a preset proportion; an alarm packet is generated to report a monitoring center and logs are recorded.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and system for verifying abnormal load data feedback from a power distribution cabinet. Background Technology

[0002] As a core component of power supply, the operation of the power distribution system directly impacts electricity safety and energy management efficiency. With the deepening of smart grid construction, real-time monitoring and remote transmission of load data from distribution cabinets have become crucial for ensuring stable grid operation. However, existing remote data acquisition solutions often focus on optimizing communication protocols or handling single-point failures, neglecting the issue of data integrity verification throughout the entire transmission chain. When network conditions fluctuate or equipment malfunctions, these solutions struggle to accurately determine the root cause of data distortion, leading to discrepancies between the load data received by the remote platform and the actual on-site conditions.

[0003] During remote transmission of load data from distribution cabinets, packet loss, out-of-order delivery, and sudden jumps in values ​​frequently occur due to factors such as network jitter, communication interference, or equipment failure. If these abnormal data are not identified in a timely manner, they will directly contaminate the load analysis results of the remote platform. For example, in the power distribution monitoring system of an industrial park, due to unstable transmission links, the locally collected load value may be 120 kW, but the value received by the remote platform may jump to 180 kW. This distorted data triggers incorrect load warnings, leading to unnecessary on-site troubleshooting by maintenance personnel. More seriously, when out-of-order data occurs, load values ​​from different times are uploaded in a disordered order, making it impossible for the remote platform to accurately reconstruct load change trends, thereby affecting the accuracy of load forecasting and dispatching decisions.

[0004] Existing solutions lack a mechanism for correlating and verifying local and remote data, making it impossible to establish a data correspondence between the acquisition and receiving ends. When the remote platform detects abnormal data, it cannot trace the corresponding local original record, nor can it determine whether the anomaly was introduced during transmission or due to a problem in the acquisition process itself. This disconnect between the two sets of data makes it difficult to accurately locate and correct abnormal data. Taking the aforementioned industrial park as an example, when maintenance personnel retrieve historical data from the remote platform, they find multiple abnormal peaks during a certain period. However, because they cannot correlate local timestamps and communication status records, they cannot determine whether these peaks are due to actual load fluctuations or transmission distortion.

[0005] Therefore, establishing an accurate correspondence between locally collected data and remotely received data during the remote transmission of load data from distribution cabinets, promptly identifying and processing abnormal transmitted data through consistency verification, and preventing distorted data from interfering with load analysis and decision-making on the remote platform have become key issues in improving the reliability of power distribution system data. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for verifying abnormal load data feedback from distribution cabinets, in order to solve the technical problem of data distortion caused by network transmission between local load measurement values ​​and remote load feedback data from distribution cabinets.

[0007] The technical solution of the present invention is as follows:

[0008] This invention provides a method for verifying abnormal load data feedback from a power distribution cabinet, mainly including:

[0009] The system acquires local load measurements from the distribution cabinet and remote load feedback data, extracts timestamps to establish a time alignment mapping table, and performs linear interpolation compensation on mismatched data points to obtain time-synchronized load data pairs. It compares the load data pairs point-by-point to calculate numerical deviations, sorts them by time to obtain a difference sequence, divides the difference sequence by time period, and extracts the maximum and average deviations for each time period as historical difference reference baselines. Based on the historical difference reference baselines, it determines the difference threshold, and uses the isolated forest algorithm to calculate the path length normalization coefficient for each value in the difference sequence to obtain an anomaly score. If the anomaly score is greater than a preset threshold and the corresponding deviation value exceeds a preset multiple of the difference threshold, it is marked as a distortion anomaly marker. It extracts the transmission delay within a preset time period before and after the distortion anomaly marker and calculates the fluctuation amplitude. If the fluctuation amplitude is greater than a preset fluctuation threshold and the peak delay exceeds a preset peak threshold, it is determined as packet loss, and the start and end timestamps are recorded to constitute an abnormal time period. Based on the abnormal time period, it generates a data retransmission request and sends it to the remote device. It receives the retransmitted load data, extracts the corresponding load values, re-inserts them according to the original timestamps, and updates the difference sequence. A classifier is used to classify the updated difference sequences, and the number of abnormal packet loss values ​​is counted. If the proportion exceeds a preset ratio, a data packet loss problem is confirmed. The distribution cabinet identification code and the abnormal timestamp are extracted and combined to form an alarm information data packet, which is sent to the monitoring center through the fault reporting channel to trigger equipment abnormality handling, and the packet loss event is recorded in the distribution cabinet fault log database.

[0010] This invention provides a power distribution cabinet load data anomaly feedback verification system, mainly comprising:

[0011] The data acquisition and time alignment module is used to acquire the local load measurement value and the remote feedback load data of the distribution cabinet, extract the timestamp to establish a time alignment mapping table, and perform linear interpolation compensation for mismatched data points to obtain time-synchronized load data pairs.

[0012] The difference calculation and baseline extraction module is used to compare load data pairs point by point to calculate the numerical deviation, sort the difference sequence by time, divide the difference sequence by time period, and extract the maximum and average deviation of each time period as historical difference reference baseline.

[0013] The anomaly detection and labeling module is used to determine the difference threshold based on the historical difference reference baseline, and to calculate the path length normalization coefficient of each value in the difference sequence using the isolated forest algorithm to obtain the anomaly score. If the anomaly score is greater than the preset threshold and the corresponding deviation value exceeds the preset multiple of the difference threshold, it is marked as a distortion anomaly marker point.

[0014] The packet loss determination and time period recording module is used to extract the transmission delay within a preset time period before and after the distortion anomaly marker and calculate the fluctuation amplitude. If the fluctuation amplitude is greater than the preset fluctuation threshold and the peak delay exceeds the preset peak threshold, it is determined to be packet loss, and the start and end timestamps are recorded to form the abnormal time period.

[0015] The data retransmission and difference update module is used to generate a data retransmission request based on the abnormal time period and send it to the remote device, receive the retransmission load data, extract the corresponding load value, re-insert it according to the original timestamp, and update the difference sequence.

[0016] The resend verification and anomaly confirmation module is used to classify the updated difference sequences using a classifier, count the number of packet loss anomaly values, and if the proportion exceeds the preset proportion, it is confirmed that there is a data packet transmission loss problem.

[0017] The alarm reporting and log recording module is used to extract the power distribution cabinet identification code and abnormal timestamp to form an alarm information data packet, which is sent to the monitoring center through the fault reporting channel to trigger equipment abnormality handling, and the packet loss event is recorded in the power distribution cabinet fault log database.

[0018] The beneficial effects of this application are as follows: A method and system for verifying abnormal load data transmission from a distribution cabinet, in use, establishes a second-level timestamp alignment mapping table by sampling at a minute frequency, uses linear interpolation to compensate for time-mismatched data points, obtains synchronized load data pairs, calculates numerical deviations point by point, and generates a difference sequence. This invention utilizes the isolated forest algorithm to construct a decision tree set, calculates the path length normalization coefficient and anomaly score of each value in the difference sequence, and combines this with a difference threshold determined by the historical seven-day average deviation baseline to accurately mark distorted anomaly points. For the marked anomaly points, this invention extracts the data packet transmission delay within 30 seconds before and after, analyzes the fluctuation amplitude and peak delay to determine packet loss anomalies and generates retransmission requests, and uses a support vector machine radial basis kernel function classifier to verify the repair effect after receiving the retransmitted data. If more than 30% of packet loss anomalies still exist, an alarm mechanism is triggered to report to the monitoring center. This invention, through closed-loop processing of time alignment, anomaly detection, intelligent retransmission, and classification verification, effectively solves the problem of packet loss and distortion in the remote transmission of load data from distribution cabinets, improving the reliability and accuracy of data acquisition. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for verifying abnormal load data feedback from a power distribution cabinet according to the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of a power distribution cabinet load data anomaly feedback verification system according to the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0024] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0025] A specific embodiment of the distribution cabinet load data anomaly feedback verification method and system of the present invention is as follows:

[0026] like Figure 1 As shown, the specific method for verifying abnormal load data feedback from a distribution cabinet in this embodiment may include:

[0027] Step S101: Obtain the local load measurement value and the remote feedback load data of the distribution cabinet. Extract the timestamps from the local load measurement value and the remote feedback load data according to the sampling frequency of one minute to obtain an initial timestamp set containing second-level time information. Based on the second-level time information in the initial timestamp set, establish a time alignment mapping table between the local load measurement value and the remote feedback load data.

[0028] For the time alignment mapping table, a traversal operation is performed to determine if there are data points with mismatched timestamps. If a mismatched data point exists, its corresponding missing value is extracted. For the missing value, a linear interpolation method is used to calculate the load slope of the adjacent time nodes. The compensation amount corresponding to the missing value is generated based on the load slope.

[0029] In this embodiment, a linear interpolation algorithm is used to restore the load data points marked as null in the mapping table. Specifically, let the standard timestamp corresponding to the missing point be... On the standard timeline, the retrieval distances forward and backward are calculated separately. The most recent valid data point is denoted as and ,in , , These are the known load measurements corresponding to these two moments.

[0030] First, calculate the slope (k) of the load change over time using the formula: In the formula, the denominator The numerator represents the time interval (in seconds) between two valid nodes. The slope (k) is the change in load within this interval, expressed in kilowatts per second (kW / s). Then, the time difference between the missing point and the forward valid point is calculated. (Unit: seconds), and generate the corresponding compensation amount. : Finally, the restored value of the missing points. The result is obtained by adding the compensation amount to the forward effective value: .

[0031] For example: Let , ; , The interval between the two valid points is... seconds, slope For missing points Its time difference with the forward point Seconds, compensation amount , restore value The calculated result Fill in the corresponding empty spaces in the mapping table to obtain complete alignment data for subsequent difference comparison.

[0032] After obtaining the compensation amount, fill the compensation amount into the corresponding timestamp position in the time alignment mapping table to obtain a complete time alignment mapping table containing the compensation amount. Based on the complete time alignment mapping table, extract the corresponding local load measurement value and remote backhaul load data to obtain a time-synchronized load data pair.

[0033] In this embodiment, if the second-level value of the local load measurement matches the second-level value of the remote load data, a time alignment mapping table is directly established. By performing linear interpolation compensation on the missing data, it is ensured that subsequent difference analysis is based on a complete and synchronized data foundation, avoiding spurious deviations caused by time misalignment.

[0034] According to an embodiment of this application, a method for establishing a time alignment mapping table is provided, as follows:

[0035] First, a standard time series slot is pre-constructed with a sampling period of minutes. This slot contains consecutive whole-minute timestamps, such as "10:00:00", "10:01:00", "10:02:00", etc., and serves as the primary key of the mapping table. Each slot corresponds to one mapping record. For each minute slot, the actual second-level timestamp recorded by the local measuring device within that minute and its corresponding load measurement value are obtained. Simultaneously, the second-level timestamps and their corresponding load data from the remote device within the same minute are also obtained. If the second value of the local timestamp is the same as the second value of the remote timestamp (e.g., 10:01:15 locally and 10:01:15 remotely), then the local timestamp, local load measurement value, remote timestamp, and remote load data are directly filled into the corresponding field of the standard slot to complete the alignment mapping.

[0036] If the second value of the local timestamp is inconsistent with the second value of the remote timestamp (e.g., local is 10:02:02, remote is 10:02:05), or if one side has no valid data within that minute, the standard slot is still retained. In this case, the actual existing timestamp and measurement value are filled into the corresponding fields. For the missing or mismatched side, both the corresponding timestamp field and the numerical field are marked as null or NaN. After traversing all minute slots, a mapping table containing a complete timeline and clearly indicating the aligned and misaligned states is generated. Subsequently, data fusion or interpolation can be performed based on the valid fields in this table.

[0037] Step S102: By comparing the local load measurement values ​​with the remote load data transmitted back in the time-synchronized load data pairs point by point, the numerical deviation at each time point is calculated. All deviation values ​​are arranged in chronological order to obtain a difference sequence containing continuous difference values.

[0038] The system acquires local load measurements after time synchronization and remote load data, compares the two values ​​point by point, calculates the numerical deviation at each time point, and obtains an initial deviation set. The Isolation Forest algorithm is then used to detect anomalies in the initial deviation set.

[0039] In one possible implementation, the isolated forest algorithm is used to perform anomaly detection on the initial set of deviations. Specifically, firstly, let the initial set of deviations be... ,in For the total number of bias samples, each element This is a single-dimensional numerical value, representing the difference between the local load value and the remote load value corresponding to a specific standard time slot. Then, from the set... A subset of samples is randomly selected. For each sample in this subset, since each sample has only one dimension ("bias value"), this dimension is directly selected as the splitting attribute. A split point is randomly chosen between the minimum and maximum values ​​of this dimension in the subset, dividing the sample into left and right subsets. This splitting process is recursively repeated for each subset until a preset tree height limit is reached, or only one sample remains in each subset. Then, the process of constructing the random decision tree is repeated, generating a predetermined number of decision trees (e.g., 100 trees), which together form an isolated forest. Finally, for the set... Each deviation value in The average path length of a sample across all trees in an isolated forest is calculated. A shorter path length indicates a more likely isolated sample and a higher degree of anomaly. An anomaly score is calculated based on this average path length, ranging from 0 to 1. A preset anomaly threshold (e.g., 0.6) is used as the criterion: if a sample's anomaly score exceeds this threshold, the sample is considered an anomaly and is marked or removed; if it does not exceed the threshold, it is considered a normal sample and retained. Through these steps, an automated anomaly screening can be performed on the initial set of deviations, eliminating abnormal deviation values ​​caused by occasional factors such as communication interference, thereby obtaining a more reliable set of deviations for subsequent analysis.

[0040] After obtaining the initial set of deviations, the Isolation Forest algorithm is used to detect anomalies in each deviation sample. The algorithm outputs an anomaly score for each sample, which measures the difficulty of isolating the sample, with a value ranging from 0 to 1. A preset anomaly detection threshold (e.g., 0.6) is set; if a sample... If the abnormal score exceeds the threshold, the deviation corresponding to the sample is determined to be a specific deviation.

[0041] If a specific deviation in the initial deviation set is identified as an outlier, that specific deviation is removed, resulting in a cleaned deviation set. The specific deviation refers to a sudden maximum value in the initial deviation set caused by sporadic non-network packet loss factors (such as strong electromagnetic interference or sudden sensor malfunctions), whose value significantly deviates from the normal deviation range. Through the automatic identification using the isolated forest algorithm, this specific deviation can be accurately located and marked without manually setting a fixed absolute deviation threshold, allowing for subsequent removal or separate processing, thus avoiding contamination of the analysis results by sporadic interference.

[0042] The elements in the cleaned deviation set are arranged chronologically to construct an initial difference sequence containing continuous values. This initial difference sequence is then segmented according to a predefined hourly division rule to obtain multiple subsequences of differences arranged by hour.

[0043] After anomaly detection and removal of specific deviations (outliers), the remaining deviation samples constitute the cleaned deviation set. Each element in this set is a deviation value that was not identified as an anomaly and accurately reflects the transmission differences between the local and remote ends of the distribution cabinet under normal operating conditions. For example, if the initial deviation set contains 60 sample points, after the Isolation Forest algorithm identifies and removes 2 outliers, the remaining 58 samples are the elements of the cleaned set.

[0044] To facilitate subsequent time-segmented analysis, this embodiment divides the cleaned deviation set into time segments according to a preset hourly division rule. The preset hourly division rule is as follows: using whole hours as boundaries and 60-minute fixed time windows, the continuous deviation sequence is cut into multiple discrete deviation subsequences along the time axis. For example, collecting local load measurements and remote feedback data for a full 24 hours, forming one deviation sample per minute, results in an initial deviation set containing 1440 minute-level sample points. Following the above division rule, using whole hours as boundaries (e.g., [00:00-01:00), [01:00-02:00), ..., [23:00-24:00)), these 1440 sample points are divided into 24 deviation subsequences. Each subsequence corresponds to 60 deviation samples within a whole hourly time segment, allowing for subsequent statistical, comparative, or trend analysis on an hourly basis.

[0045] By traversing each differential subsequence, the maximum and average deviation values ​​in each differential subsequence are extracted to generate a baseline feature vector. Here, the deviation refers to the absolute difference between the local load measurement and the remote transmitted load data at the same time slot alignment. The baseline feature vector is in the form of [maximum deviation value, average deviation value].

[0046] If the maximum deviation value in the baseline feature vector is greater than a preset deviation threshold, the baseline feature vector is smoothed to obtain the target feature vector. The target feature vector is then determined as the historical difference reference baseline.

[0047] In one possible implementation, after generating the baseline feature vectors corresponding to each hourly time period, the vectors are further filtered and corrected. Specifically: First, a preset deviation threshold is set. This threshold is used to determine whether the maximum deviation value in the baseline feature vectors of each time period is within the normal range. This threshold is set according to the accuracy level of the distribution cabinet equipment or the maximum normal ripple error allowed by the communication protocol. For example, if the rated power of the distribution cabinet is 200kW, and the normal error fluctuation allowed for remote transmission is within ±2% of the rated value (i.e., 4kW), and a margin of 2.5 times is taken based on historical experience, then the preset deviation threshold is set to 2.5×4=10kW. Subsequently, the baseline feature vectors v=[maximum deviation value, average deviation value] of each time period are traversed, and it is checked whether the maximum deviation value of its first dimension is greater than the above-mentioned preset deviation threshold. If the maximum deviation value of a certain time period does not exceed the threshold, then the baseline feature vector of that time period is directly determined as the target feature vector, as the historical difference reference baseline for that time period. If the maximum deviation value of a certain time period (denoted as the t-th hour) exceeds the preset deviation threshold, for example, the baseline feature vector of that time period... If the maximum deviation value of 15kW is greater than the threshold of 10kW, smoothing processing is triggered.

[0048] In one possible implementation, the smoothing process employs a mean substitution method for neighboring time periods: obtaining the first... Hourly feature vector With the Hourly feature vector Calculate the arithmetic mean of the two, and replace the original outlier vector with this mean. The formula is: In this case, vector addition involves adding the corresponding dimensions separately, while division involves dividing each dimension by 2. For example, if... , ,but: Using the smoothed vector Replace the original vector The corrected vector is then identified as the target feature vector and used as the historical difference reference baseline for that time period. By processing all time periods in the above manner, a stable and reliable historical difference reference baseline for each time period is finally obtained for subsequent packet loss boundary determination.

[0049] In this embodiment, the differential sequence is segmented according to the hourly dimension, which can effectively capture the differential characteristics of load data at different time periods. By extracting the maximum deviation value and the average deviation value as historical difference reference baselines, a reliable benchmark is provided for subsequent anomaly judgment.

[0050] As an optional embodiment of this application, the initial deviation set is cleaned using the Isolation Forest algorithm, which can effectively identify and remove extreme deviation points caused by occasional interference. The cleaned data more accurately reflects the actual difference between the local and remote ends of the distribution cabinet, avoiding misjudgment.

[0051] Step S103: Determine the difference threshold based on the average deviation value over seven consecutive days in the historical difference reference baseline. Calculate the path length normalization coefficient for each value in the difference sequence using the Isolation Forest algorithm, and generate anomaly scores based on this coefficient. If the anomaly score is greater than a preset threshold and the corresponding deviation value exceeds a preset multiple (preferably twice) of the difference threshold, then extract the time point corresponding to that value as a distortion anomaly marker.

[0052] Historical deviation reference baseline data for seven consecutive days was obtained. Each day was divided into several whole-hour periods, and the average deviation value within each period was calculated. The arithmetic mean of the average deviation values ​​for all periods was taken as the deviation threshold. The unit of the deviation threshold is kilowatts, representing a benchmark value for the normal fluctuation level of the network.

[0053] Specifically, let there be N effective hour periods over seven consecutive days, and the historical average deviation value for the i-th period is... Then the difference threshold for:

[0054]

[0055] Where N is the total number of valid time periods within seven consecutive days, and the average deviation value is... The deviation value for all standard time slots within this period. The arithmetic mean of the values, with the deviation value in kilowatts (kW).

[0056] For the differential sequence data to be processed (i.e., a one-dimensional numerical sequence composed of deviation values ​​under each standard time slot), the Isolation Forest algorithm is used to construct an Isolation Forest model containing a predetermined number of isolated trees. During construction, each tree randomly selects a predetermined number of samples from the differential sequence as a subset. Since each sample has only one feature dimension, "deviation value," a split point is randomly selected within the range of values ​​for this dimension during construction, recursively dividing the samples into left and right subsets until the predetermined tree height limit is reached or only one sample remains in the subset. All isolated trees together constitute the Isolation Forest model.

[0057] For each deviation value in the discrepancy sequence, its average path length across all isolated trees in the isolated forest model is calculated; that is, the average number of edges traversed from the root node to the isolated tree on each isolated tree. Based on the average path length, a dimensionless outlier score is calculated using the path length normalization formula. This score ranges from 0 to 1; the closer the score is to 1, the more likely the sample is to be isolated, and the higher the degree of outlier.

[0058] Specifically, let the deviation value of a sample be x, and its path length on a certain isolated tree be h(x). Then, its average path length across all trees is E(h(x)). The formula for calculating the anomaly score s(x) is: .in, For the size of a single isolated tree subsample, For inclusion The baseline value for the average path length of the binary search tree for each sample. The outlier score is a dimensionless value.

[0059] When determining distortion anomaly markers, a dual verification logic is used: if the anomaly score of a sample is greater than the preset anomaly threshold, and the actual physical deviation value corresponding to the sample is greater than the preset multiple of the difference threshold, then when both conditions are met, the time point corresponding to the value is extracted as the distortion anomaly marker.

[0060] Specifically, let the preset abnormal threshold be... (For example, 0.6 is used in industrial practice), the preset multiple of the difference threshold is (For example, take 2 times). If and only if the following condition is met. and At that time, the time point corresponding to the sample is marked as a distortion anomaly marker point.

[0061] Based on the distortion anomaly markers, adjacent data segments at corresponding time positions in the difference sequence are extracted to obtain anomalous data fragments for subsequent troubleshooting or data correction. This dual verification utilizes the statistical sensitivity of isolated forests to identify anomalies while also using actual physical deviation values ​​to avoid logical errors caused by directly comparing dimensionless scores with dimensional thresholds.

[0062] In this embodiment, the average deviation over seven consecutive days effectively smooths out daily fluctuations, making the difference threshold more representative. By using the path length normalization calculation of the Isolation Forest algorithm, outliers deviating from the normal range can be accurately identified, providing accurate labeling criteria for subsequent packet loss determination.

[0063] As an optional embodiment of this application, the number of isolated trees is set to one hundred, which ensures both detection accuracy and computational efficiency. When the anomaly score of a sample is greater than the preset anomaly threshold, and the actual physical deviation value corresponding to the sample is greater than twice the difference threshold, it indicates that the data point has significant abnormal characteristics and its cause needs further analysis.

[0064] Step S104: For the distortion anomaly marker, obtain the network latency data recorded by the distribution cabinet communication module. Extract the data packet transmission latency within a preset time period before and after the distortion anomaly marker. Calculate the fluctuation range between the peak and trough values ​​of the transmission latency within this time period. If the fluctuation range is greater than a preset fluctuation threshold and the peak latency exceeds a preset peak threshold, then the anomaly type corresponding to the distortion anomaly marker is determined to be data packet transmission loss. Record the start and end timestamps of the loss occurrence to constitute the abnormal time period.

[0065] For each distortion anomaly marker, the network latency is obtained from the distribution cabinet communication terminal. The first transmission segment within a preset time period (e.g., 30 seconds) before and after the distortion anomaly marker is obtained through time window interception.

[0066] The network latency is collected through the network layer or application layer message exchange mechanism of the power distribution cabinet communication module. Specifically, it can be obtained using either of the following methods: Method 1 (Network Layer): The background daemon of the communication module sends network probe request packets to the remote acquisition platform at a preset frequency and records the time difference between receiving the response packet. This time difference is used as the current physical network round-trip time. Method 2 (Application Layer): The power distribution cabinet sends a load data packet containing a local sending timestamp to the remote end. After receiving the packet, the remote end returns a response packet containing a remote receiving timestamp and an acknowledgment timestamp. The power distribution cabinet records the local receiving timestamp when receiving the response. Based on the above four timestamps, the network latency is calculated using a known round-trip time estimation formula: .in, The local timestamp for sending load data packets to the distribution cabinet. For the local timestamp of the remote response packet received by the distribution cabinet, The timestamp recorded when the payload data packet is received at the remote end. This is the timestamp recorded when the remote end sends the response packet; the formula estimates the one-way network transmission delay by deducting the processing time consumed by the remote end, in milliseconds (ms).

[0067] Furthermore, determine the timestamps corresponding to the distortion anomaly markers. Centered on this timestamp, a preset duration (e.g., 30 seconds) is taken both forward and backward as a time window, i.e., the time frame from... Instant The time interval is specified in seconds. All network latency records falling within this time window are retrieved from the network latency history database cached locally in the distribution cabinet. This set of local latency data is defined as the first transmission segment. For example, if the distortion anomaly marker is 10:15:00, a ±30 second time window is set centered on this time. All network latency records between 10:14:30 and 10:15:30 are extracted to form the first transmission segment. This segment is then used for subsequent correlation analysis with load deviation data within this time period to determine whether the anomaly is caused by network transmission latency fluctuations.

[0068] For the first transmission segment, an outlier removal process using the Isolation Forest algorithm is employed to obtain the second transmission segment. Specifically, all network latency values ​​contained in the first transmission segment are used as the input dataset, which is a single-dimensional numerical sequence (in milliseconds). The Isolation Forest algorithm is used to detect anomalies in this dataset. Due to the small sample size in the first transmission segment (corresponding to latency records within a 60-second time window), the number of isolation trees is set to a preset lightweight value (e.g., 50 to 100 trees). When constructing each tree, a recursive binary split is performed by randomly selecting a split point between the minimum and maximum latency values ​​until a preset tree height limit is reached or the subset cannot be further divided. An anomaly score is calculated for each latency sample in the isolation forest. This score is a dimensionless value ranging from 0 to 1; the closer the value is to 1, the more likely the sample is to be isolated, and the higher the degree of anomaly. Latency samples with anomaly scores greater than a preset outlier threshold are identified as occasional outliers and removed from the first transmission segment; normal latency samples with anomaly scores less than or equal to the preset outlier threshold are retained. After the elimination is completed, the remaining normal network latency data sequence is the second transmission segment, which is used for accurate assessment of subsequent network jitter amplitude.

[0069] Obtain the data packet delay value within the second transmission segment, and determine the peak and trough points of this delay value through extreme value search and comparison. Calculate the difference between the peak and trough points to obtain the fluctuation difference.

[0070] In one possible implementation, after obtaining the second transmission segment, extreme value features are further extracted from the network latency during that period, as follows:

[0071] Obtain all network delay samples within the second transmission segment, and perform a maximum / minimum value search and comparison on the delay sequence using a single linear traversal method. Specifically, use the first delay value in the sequence as the initial peak and initial valley value, and traverse each delay value in the sequence sequentially. If the current delay value is greater than the recorded peak value, update the peak value; if the current delay value is less than the recorded valley value, update the valley value. After the traversal is complete, lock the peak and valley points within that time period. Let the peak point be... Valley point is Both are network latency parameters, measured in milliseconds (ms). The fluctuation difference is obtained by calculating the difference between the peak and trough values. : This fluctuation difference reflects the instantaneous fluctuation amplitude of network delay within the second transmission segment, measured in milliseconds. For example, suppose the delay sequence of the second transmission segment is... (Unit: milliseconds) After the above extreme value search and comparison, the peak point is locked. Valley point Then the fluctuation difference The fluctuation difference can be further compared with a preset fluctuation threshold to determine whether the distortion anomaly is caused by network packet loss or severe jitter.

[0072] If the fluctuation difference is greater than a preset fluctuation threshold and the peak point is greater than a preset peak threshold, the anomaly type corresponding to the distortion anomaly marker is determined to be data packet transmission loss. For data packet transmission loss, the start and end points of the loss are extracted, and the start and end points are combined to form an abnormal time period. The preset fluctuation threshold is set according to the normal latency jitter range specified by the power distribution cabinet communication module (e.g., 100 milliseconds), and the preset peak threshold is set according to the maximum latency tolerance allowed by the network transmission protocol (e.g., 500 milliseconds).

[0073] In this embodiment, by analyzing network latency data before and after the distortion anomaly marker, it is possible to accurately determine whether the anomaly was caused by packet loss. The 30-second time window can cover the transient process of network fluctuations while avoiding the introduction of too much irrelevant data.

[0074] As an optional embodiment of this application, the preset fluctuation threshold is set to 100 milliseconds, and the preset peak threshold is set to 500 milliseconds. These two thresholds are determined based on the normal transmission characteristics of the power distribution cabinet communication system and can effectively distinguish between normal fluctuations and abnormal packet loss.

[0075] Step S105: Based on the start and end timestamps of the abnormal time period, generate a data retransmission request command containing time range parameters. Send the data retransmission request command to the remote data acquisition device through the distribution cabinet communication interface. Receive the retransmitted load data returned by the remote data acquisition device as the second remote feedback. Extract the load value corresponding to the abnormal time period from the second remote feedback. Reinsert the load value into the time-synchronized load data pair according to the original timestamp, and update the difference sequence.

[0076] Based on the start and end timestamps of the abnormal time period, a data retransmission request command containing time range parameters is generated. This data retransmission request command is sent to a remote data acquisition device via the distribution cabinet communication interface, and the retransmission load data returned by the remote data acquisition device is obtained.

[0077] The reissue load data is acquired, and a decision tree algorithm is used to determine the integrity of the reissue load data to obtain an integrity determination result. If the integrity determination result is complete, the load value corresponding to the abnormal time period is extracted from the reissue load data.

[0078] In this embodiment, the root node of the decision tree performs sample size verification: obtaining the number of data points for reissued load data. And compare it with the total number of points that should theoretically exist within the abnormal time period. Comparison, among which It is calculated based on the duration of the abnormal time period and the sampling frequency. If If the two results are equal, the integrity result is directly determined to be incomplete; if they are equal, the process proceeds to the next level of child nodes.

[0079] The child node performs a timestamp continuity check: it checks whether the timestamp records of the reissued load data completely cover the start and end times of the abnormal time period, and whether the interval between adjacent timestamps is equal to the preset sampling period. If there are missing or discontinuous timestamps, it is determined to be incomplete; if there are no discontinuous timestamps, it proceeds to the next child node.

[0080] The next-level child node performs format and null value checks: it checks whether there are null values, non-numeric values ​​(NaN), or dead values ​​with abnormal physical meaning (such as negative power values) in the reissued load data. If any of the above exists, it is determined to be incomplete; if all checks pass, the final result is determined to be complete.

[0081] If the integrity determination result is complete, the load value corresponding to the abnormal time period in the reissue load data is extracted and used to fill the corresponding missing position in the mapping table.

[0082] If the integrity determination result is incomplete, a limited retransmission mechanism is triggered. Specifically, the system records the current retransmission count. If it does not exceed the preset maximum retransmission count (e.g., 3 times), a retransmission request is sent to the remote end. If the data is still determined to be incomplete after exceeding the maximum retransmission count, the batch of retransmission data is discarded, the corresponding missing field in the mapping table remains unchanged, the data segment is marked as quality unqualified and transferred to the subsequent processing flow for confirmation of excessive packet loss. At the same time, an equipment anomaly alarm message is sent to the monitoring center to trigger manual or advanced maintenance intervention.

[0083] The load values ​​are obtained, and then re-inserted into the synchronized load data pairs according to the original timestamps. The differences between local load measurements and remote load data at each standard time slot are recalculated to obtain the first update difference sequence (i.e., the load difference update sequence). Based on the first update difference sequence, a support vector machine algorithm is used for classification to determine the final load status category.

[0084] In one possible implementation, after updating the mapping table by retransmitting load data, a pre-trained support vector machine (SVM) model is further used to determine the final load state category based on the first update difference sequence. The specific implementation is as follows:

[0085] The first updated difference sequence As the input feature vector, where m is the total number of sample points in the differential sequence, each element... This represents the deviation between the local load measurement and the remote load data at the corresponding standard time slot, expressed in kilowatts (kW). Since interference factors in actual transmission environments often exhibit nonlinear characteristics, the SVM classifier uses a radial basis function kernel to map the low-dimensional deviation features to a high-dimensional linearly separable space. Using pre-trained hyperplane parameters (weight vector w and bias b), for each deviation value... Calculate the classification decision value: .in, The kernel function will The vector mapped to a high-dimensional feature space. Let b be the transpose of the weight vector, and b be the bias term. The decision value is... It is a dimensionless scalar. If If so, the deviation value is determined to belong to the "packet loss anomaly" category; if If the deviation value is within the normal or acceptable error range, then it is determined that the deviation value is within the normal or acceptable error range.

[0086] Based on the SVM classification results and the data re-upload verification, the following three final load status categories are determined: The first category is the normal load status. This occurs when all deviation values ​​in the first update difference sequence meet the following conditions. When the resent data passes integrity verification and is successfully inserted, the current load status is determined to be normal, indicating that the remote data transmission has accurately restored the actual load change trend on site. The second category is the packet loss delay unrepaired state. When the first update difference sequence is determined by SVM to be "packet loss anomaly" (i.e. When the proportion of sample points in a given period exceeds a preset threshold (e.g., 30%), the current load status is determined to be packet loss and delay unresolved, indicating that there is still a serious packet loss problem in the communication link. The third category is the actual load fluctuation status. When the consistency check between the local load measurement value and the remote return data is completely consistent, and the deviation values ​​are all within the allowable range of system error, the load peak recorded by the remote platform is determined to be the actual load fluctuation that occurred locally, rather than an illusion caused by transmission distortion. After determining the final load status category, the category label is associated with the corresponding time period and recorded in the system log for subsequent operation and maintenance decisions or alarm triggering.

[0087] In this embodiment, data loss caused by network packet loss can be filled by requesting resend data from a remote data acquisition device. The integrity determination of the decision tree algorithm ensures the reliability of the resend data quality and avoids new anomalies caused by replacing the original data with incomplete data.

[0088] As an optional embodiment of this application, the data retransmission request command includes a precise time range parameter, enabling the remote device to accurately locate the data segment that needs to be retransmitted. This precise request method reduces unnecessary data transmission and improves communication efficiency.

[0089] Step S106: Construct a radial basis function (RBF) classifier using a support vector machine. The updated difference sequence is used as feature input, and a pre-trained classifier model is used to calculate the classification decision function value for each difference value. If the decision function value is less than a preset decision threshold, the difference value is determined to still belong to the packet loss anomaly category. The number of values ​​in the updated difference sequence that are determined to be packet loss anomalies is counted. If the ratio of the number of anomaly values ​​to the total number of all data points is greater than a preset ratio, it is confirmed that there is still a data packet loss problem in the second remote transmission.

[0090] Obtain the original set of transmission packets from the target return end. Extract the original set of transmission packets according to a preset abnormal time period (i.e., the abnormal time period constituted in S104) to obtain the first transmission packet within the target time period. For the first transmission packet, extract all data points from it, and construct an initial sequence set based on all data points.

[0091] Obtain adjacent data points in the initial sequence set, calculate the difference value between the adjacent data points, and generate a second updated difference sequence set (i.e., communication waveform difference sequence set) based on the difference value of each adjacent data point.

[0092] The second updated difference sequence set is obtained as a feature and input into a pre-trained support vector machine (SVM) radial basis function (RBF) classifier. The SVM RBF classifier calculates a classification decision value for each difference value in the second updated difference sequence set. The classification decision value is obtained; if the classification decision value is less than a preset decision threshold, the difference value is determined to belong to the packet loss anomaly class.

[0093] In this embodiment, a pre-trained Support Vector Machine (SVM) classifier is used to identify packet loss anomalies. The training process is as follows: Historical transmission segments are collected and manually labeled with category tags. Segments that transmit normally without packet loss are labeled as +1 (normal class), and segments that cause waveform breaks or numerical jumps due to network congestion or intermittent packet loss are labeled as -1 (packet loss anomaly class). The differences between adjacent data points in each segment are extracted as features to construct the input feature vector. A radial basis function kernel is used to map the features to a high-dimensional space. The support vectors, weight coefficients, biases, and kernel parameters are obtained through optimization, completing the classifier training.

[0094] In actual classification, the second updated difference sequence set is used as input features and fed into a pre-trained support vector machine radial basis function classifier. The classifier classifies each difference value in the second updated difference sequence set... Using the parameters obtained from training and the RBF kernel function, calculate its classification decision value: .in, The feature vector representing the current input difference values. For the set of support vectors, For the first The weight coefficients of each support vector. For the first Each support vector corresponds to a category label (with a value of +1 or -1). For radial basis kernel functions, For bias terms, The output is the classification decision value, which is a dimensionless real number representing the direction and distance of the current difference value from the classification hyperplane.

[0095] Obtain classification decision value Then, it is compared with a preset decision threshold. In this embodiment, the preset decision threshold is preferably -1, which is the standard value of the margin boundary where the support vectors are located. Using this threshold as the decision boundary is a conservative robustness decision strategy. If the difference value is determined to belong to the packet loss exception category; if If the difference value is not found to be normal, it is determined to belong to the category of normal network fluctuations or normal transmission that has been repaired, and it is not included in the packet loss statistics. After classifying each difference value in the second updated difference sequence set in the above manner, the number of difference values ​​belonging to the packet loss anomaly category is counted for subsequent load status category determination.

[0096] The number of abnormal values ​​corresponding to the packet loss anomaly category is counted. If the ratio of the number of abnormal values ​​to the total number of all data points is greater than a preset proportion, it is determined that the target backhaul end has an excessive packet loss problem. The preset proportion is determined by the minimum service tolerance of the power distribution network for data integrity, and is preferably 30%. In this embodiment, a secondary verification of the retransmitted data using a support vector machine radial basis function classifier can effectively identify whether packet loss still exists in the retransmitted data. A decision function value less than -1 indicates that the data point has obvious abnormal characteristics and should be classified into the packet loss anomaly category.

[0097] As an optional embodiment of this application, the preset proportion is set to 30%. When the proportion of abnormal values ​​exceeds this threshold, it indicates that the quality of the resubmitted data is poor and still cannot meet the requirements of normal analysis, requiring further abnormality handling procedures to be triggered.

[0098] As another optional embodiment of this application, the radial basis function kernel classifier is trained with a large amount of historical packet loss data and can accurately identify various packet loss patterns. This classifier has good discrimination ability for both sudden and persistent packet loss, ensuring the reliability of anomaly detection.

[0099] Step S107: To address the data packet loss issue that still exists in the second remote backhaul, extract the distribution cabinet identification code and the start timestamp of the abnormal time period, and combine them to form an alarm information data packet. Send the alarm information data packet to the monitoring center through the distribution cabinet's fault reporting channel, triggering the monitoring center's equipment anomaly handling process, and recording this packet loss anomaly event in the distribution cabinet fault log database.

[0100] The first data packet returned from the remote endpoint is acquired, and a packet loss detection tool is used to detect it, resulting in a second data packet with a loss amount greater than a preset threshold. Specifically, the sequence number of each packet is read sequentially. If the difference between the sequence number of the current packet and the sequence number of the previous packet is greater than 1, it is determined that there is a lost packet between them. The total number of lost packets in the first data packet is counted to obtain the loss amount. The loss amount is compared with a preset threshold. If the loss amount is greater than the preset threshold, the first data packet is identified as the second data packet. The first data packet contains multiple load data packets, and the header of each packet carries a continuously increasing sequence number. The packet loss detection tool is a network protocol analysis component or a packet sequence number verifier. The preset threshold is pre-set based on the transmission frequency of the distribution cabinet and the network baseline packet loss rate.

[0101] For example, if the theoretical number of packets to be received in a certain minute is 60, and the packet loss detection tool detects that the actual number of received packets is 53, then the loss is 7 packets. If the preset threshold is 5, since 7 is greater than 5, the packet is identified as the second packet for use in the subsequent fault characterization process.

[0102] The identifier and timestamp corresponding to the second data packet are obtained, and a first alarm packet is generated by combining the identifier and timestamp. The first alarm packet is sent to the monitoring terminal through the reporting terminal, and an anomaly detection is performed on the first alarm packet using the isolated forest algorithm to obtain a second alarm packet with an anomaly score greater than a preset score. Each alarm packet in the first alarm packet set contains at least two feature dimensions: alarm time interval and data packet size.

[0103] In one possible implementation, the isolated forest algorithm is used to detect anomalies in the first alarm packet. Specifically: the alarm time interval and data packet size of each alarm packet are extracted from the first alarm packet set to construct a two-dimensional feature space; for this feature space, a feature dimension is randomly selected, and a split point is randomly selected between the maximum and minimum values ​​of this dimension to perform binary tree partitioning, recursively constructing a predetermined number of isolated trees (e.g., 100 trees) to form an isolated forest model; the average path length of each alarm packet in the isolated forest is calculated, and an anomaly score is calculated based on the average path length. The anomaly score is a dimensionless value, ranging from 0 to 1. The closer the value is to 1, the easier it is for the alarm packet to be isolated, and the higher the degree of anomaly. The anomaly score of each alarm packet is compared with a preset score. If the anomaly score of an alarm packet is greater than the preset score, the alarm packet is determined to be a valid abnormal alarm packet and is extracted as a second alarm packet. The preset score is pre-set based on the tolerance for false alarm rate, with a preferred value of 0.65. For example, if the abnormal score of a false alarm packet is calculated to be 0.78, while the preset score is 0.65, since 0.78 is greater than 0.65, the alarm packet is extracted as a second alarm packet for use in subsequent high-risk alarm processing procedures.

[0104] The monitoring terminal is triggered by the second alarm packet to perform an abnormal process. Device status information is extracted from the abnormal process to obtain first device status information. A hierarchical clustering algorithm is used to cluster the first device status information to determine second device status information belonging to the fault category.

[0105] In one possible implementation, a hierarchical clustering algorithm is used to cluster the first device status information. The specific steps are as follows: First, initialize the clusters: treat each sample point in the first device status information as an independent initial cluster. Then, calculate the inter-cluster distance: use Euclidean distance to calculate the distance between the feature vectors of different sample points, and use Ward's sum of squared deviations method to calculate the distance between clusters. For any two sample points... and Its Euclidean distance for: .in, The total dimension of the feature vector. and Two samples respectively at the 1st The values ​​are taken in each feature dimension. Then, iterative merging is performed: in each iteration, the two closest clusters are selected and merged into a new cluster. Finally, truncation and qualitative analysis are performed: the iterative merging process is repeated until all samples are merged into a clustering dendrogram. The dendrogram is truncated according to a preset truncation height, dividing it into multiple categories. Categories that deviate from the normal baseline and contain hardware error codes are identified as fault categories. All state data contained in these fault categories constitute the second device state information.

[0106] If the status information of the second device meets the preset fault conditions, a packet loss exception event is generated. The packet loss exception event is written into the record table of the fault database to complete the archiving and storage of exception events.

[0107] The preset fault conditions consist of multiple limit-crossing indicators, specifically including at least one of the following conditions: communication hard fault conditions, including signal strength continuously lower than a preset strength threshold, or bit error rate continuously higher than a preset bit error rate threshold; electrical hard fault conditions, including the internal ambient temperature of the equipment exceeding a preset temperature threshold, or the three-phase voltage imbalance exceeding a preset imbalance threshold; and system log hard fault conditions, including the appearance of a preset core deadlock error code in the log.

[0108] In one possible implementation, if the status information of the second device meets any one or more of the preset fault conditions, it is determined that the power distribution cabinet has experienced a physical hardware failure or a serious environmental anomaly, and a packet loss anomaly event is generated. The packet loss anomaly event is written into a record table in the fault database. The record table includes at least the occurrence time of the anomaly event, the fault type identifier, and the corresponding status information of the second device, thus completing the archiving and storage of the anomaly event for subsequent review by maintenance personnel or automatic dispatch triggering.

[0109] In this embodiment, if resending data still fails to resolve the packet loss issue, the abnormal situation needs to be reported to the monitoring center promptly. The alarm information data packet contains the distribution cabinet identification code and the abnormal time period information, enabling the monitoring center to quickly locate the problematic equipment and the problematic time period.

[0110] As an optional embodiment of this application, the fault reporting channel uses an independent communication link to ensure that anomalies can still be reported normally when the main communication link fails. The packet loss anomaly event log contains detailed timestamps, device identifiers, anomaly types, and other information, providing a complete basis for subsequent fault analysis and equipment maintenance.

[0111] As another optional embodiment of this application, the hierarchical clustering algorithm can group similar device status information into the same category, facilitating the identification of systemic failure modes. When the second device status information meets the preset failure conditions, it indicates that the anomaly has reached a severity level requiring manual intervention, and the monitoring center should immediately activate the corresponding handling plan.

[0112] This application provides an embodiment that enables full-process monitoring and processing of discrepancies between local load data and remote feedback data from distribution cabinets. It establishes a complete closed-loop management mechanism, encompassing time synchronization, discrepancy analysis, anomaly detection, data retransmission verification, and fault reporting. This method effectively identifies and handles data packet loss issues, ensuring the accuracy and integrity of distribution cabinet load data and providing a reliable guarantee for the stable operation of the power system.

[0113] like Figure 2 As shown, this invention provides a power distribution cabinet load data anomaly feedback verification system, mainly comprising:

[0114] The data acquisition and time alignment module is used to acquire the local load measurement value and the remote feedback load data of the distribution cabinet, extract the timestamp to establish a time alignment mapping table, and perform linear interpolation compensation for mismatched data points to obtain time-synchronized load data pairs.

[0115] The difference calculation and baseline extraction module is used to compare load data pairs point by point to calculate the numerical deviation, sort the difference sequence by time, divide the difference sequence by time period, and extract the maximum and average deviation of each time period as historical difference reference baseline.

[0116] The anomaly detection and labeling module is used to determine the difference threshold based on the historical difference reference baseline, and to calculate the path length normalization coefficient of each value in the difference sequence using the isolated forest algorithm to obtain the anomaly score. If the anomaly score is greater than the preset threshold and the corresponding deviation value exceeds the preset multiple of the difference threshold, it is marked as a distortion anomaly marker point.

[0117] The packet loss determination and time period recording module is used to extract the transmission delay within a preset time period before and after the distortion anomaly marker and calculate the fluctuation amplitude. If the fluctuation amplitude is greater than the preset fluctuation threshold and the peak delay exceeds the preset peak threshold, it is determined to be packet loss, and the start and end timestamps are recorded to form the abnormal time period.

[0118] The data retransmission and difference update module is used to generate a data retransmission request based on the abnormal time period and send it to the remote device, receive the retransmission load data, extract the corresponding load value, re-insert it according to the original timestamp, and update the difference sequence.

[0119] The resend verification and anomaly confirmation module is used to classify the updated difference sequences using a classifier, count the number of packet loss anomaly values, and if the proportion exceeds the preset proportion, it is confirmed that there is a data packet transmission loss problem.

[0120] The alarm reporting and log recording module is used to extract the power distribution cabinet identification code and abnormal timestamp to form an alarm information data packet, which is sent to the monitoring center through the fault reporting channel to trigger equipment abnormality handling, and the packet loss event is recorded in the power distribution cabinet fault log database.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The scope of patent protection of the present invention shall be determined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention shall also be included within the scope of protection of the present invention.

Claims

1. A method for verifying abnormal load data feedback from a power distribution cabinet, characterized in that, The method includes: Acquire the local load measurement value and the remote feedback load data of the distribution cabinet, extract the timestamp to establish a time alignment mapping table, perform linear interpolation compensation for mismatched data points, and obtain time-synchronized load data pairs; The load data were compared point by point to calculate the numerical deviation. The difference sequence was sorted by time to obtain the difference sequence. The difference sequence was divided into time periods and the maximum and average deviations of each time period were extracted as historical difference reference baselines. The difference threshold is determined based on the historical difference reference baseline. The path length normalization coefficient of each value in the difference sequence is calculated using the isolated forest algorithm to obtain the anomaly score. If the anomaly score is greater than the preset threshold and the corresponding deviation value exceeds the preset multiple of the difference threshold, it is marked as a distortion anomaly marker point. Extract the transmission delay within a preset time period before and after the distortion anomaly marker and calculate the fluctuation amplitude. If the fluctuation amplitude is greater than the preset fluctuation threshold and the peak delay exceeds the preset peak threshold, it is determined to be packet loss, and the start and end timestamps are recorded to form an abnormal time period. Data retransmission requests are generated based on the abnormal time period and sent to the remote device. The retransmission load data is received, the corresponding load values ​​are extracted, and the data is re-inserted according to the original timestamp and the difference sequence is updated. A classifier is used to classify the updated difference sequences, and the number of values ​​for packet loss anomalies is counted. If the proportion exceeds the preset proportion, it is confirmed that there is a data packet transmission loss problem. The power distribution cabinet identification code and abnormal timestamp are extracted and combined to form an alarm information data packet, which is sent to the monitoring center through the fault reporting channel to trigger equipment abnormality handling, and the packet loss event is recorded in the power distribution cabinet fault log database.

2. The method according to claim 1, characterized in that, The process of extracting timestamps to establish a time alignment mapping table, and performing linear interpolation compensation on mismatched data points to obtain time synchronization load data pairs, specifically includes: The standard time series slots are constructed using the sampling period as the primary key of the mapping table; For each time slot, obtain the second-level timestamp and load value of the local measurement value and the remote transmission data; If the local second-level timestamp is consistent with the remote second-level timestamp, then directly fill in the corresponding field to complete the alignment; If the second-level timestamps are inconsistent or there is no valid data, the corresponding field will be marked as null. For data points marked as null, extract the effective load values ​​of adjacent time nodes, calculate the load slope, and generate compensation amounts to fill in the mapping table.

3. The method according to claim 1, characterized in that, The step of dividing the differential sequence into time periods and extracting the maximum and average deviations of each time period as historical differential reference baselines specifically includes: By comparing the local load measurement values ​​after time synchronization with the load data transmitted from the remote end point by point, the numerical deviation at each time point is calculated to obtain the initial deviation set. The isolated forest algorithm is used to perform anomaly detection on the initial set of deviations, and the deviations that are identified as anomalies are removed to obtain a cleaned set of deviations. Arrange the elements in the set of post-cleaning deviations in chronological order to construct an initial difference sequence; The initial difference sequence is divided according to a preset time period division rule to obtain multiple difference subsequences; The maximum and average deviation values ​​of each differential subsequence are calculated to generate a baseline feature vector. If the maximum deviation value in the baseline feature vector is greater than a preset deviation threshold, then smoothing is performed to obtain a historical difference reference baseline.

4. The method according to claim 1, characterized in that, The method employs the isolated forest algorithm to calculate the path length normalization coefficient of each value in the difference sequence to obtain an anomaly score. If the anomaly score is greater than a preset threshold and the corresponding deviation value exceeds a preset multiple of the difference threshold, it is marked as a distortion anomaly marker point. Specifically, this includes: The difference threshold is determined based on the average deviation over multiple consecutive days in the historical difference reference baseline; Construct an isolated forest of one hundred decision trees; Calculate the path length normalization coefficient for each value in the differential sequence and obtain the outlier score; If the anomaly score is greater than 0.6 and the corresponding deviation value exceeds twice the difference threshold, then the corresponding time point is extracted as a distortion anomaly marker.

5. The method according to claim 1, characterized in that, The transmission delay within a preset time period before and after the extracted distortion anomaly marker is calculated, and the fluctuation amplitude is calculated. If the fluctuation amplitude is greater than a preset fluctuation threshold and the peak delay exceeds a preset peak threshold, it is determined to be packet loss. The start and end timestamps are recorded to form the abnormal time period, which specifically includes: Obtain network latency from distorted anomaly markers; Extract transmission delay data within 30 seconds before and after the distortion anomaly marker; Remove outliers from transmission delay data; Determine the peak and trough values ​​of transmission delay and calculate the fluctuation amplitude; If the fluctuation range is greater than 100 milliseconds and the peak delay exceeds 500 milliseconds, it is determined that the data packet transmission is lost. Extract the start and end timestamps of the packet loss to form the abnormal time period.

6. The method according to claim 1, characterized in that, The process of generating a data retransmission request based on the abnormal time period and sending it to the remote device, receiving the retransmitted load data, extracting the corresponding load value, re-inserting it according to the original timestamp, and updating the difference sequence specifically includes: Generate a data retransmission request containing time range parameters based on the start and end timestamps of the abnormal time period; Send a data retransmission request to the remote data acquisition device through the power distribution cabinet communication interface; Receive retransmission payload data returned by the remote device and perform integrity determination; If the integrity determination result is complete, then extract the load value corresponding to the abnormal time period from the reissue load data; Reinsert the load values ​​into the time-synchronized load data pair according to the original timestamps, and update the difference sequence.

7. The method according to claim 6, characterized in that, The integrity determination specifically includes: Verify whether the number of data points for the reissued load data is consistent with the total number of data points that should theoretically exist during the abnormal time period; Verify whether the timestamps of the resent load data continuously cover the start and end times of the abnormal time period and the interval is equal to the preset sampling period; Verify that there are no null values, non-numerical values, or physically inconsistent values ​​in the reissued load data; If all the above checks pass, the data is considered complete; otherwise, it is considered incomplete and a limited number of retransmissions are triggered.

8. The method according to claim 1, characterized in that, The updated difference sequences are classified using a classifier, and the number of abnormal packet loss categories is counted. If the proportion exceeds a preset ratio, a data packet loss problem is confirmed. Specifically, this includes: Retrieve transmission packets from the remote end and extract them according to abnormal time periods; Extract all data points from the transmission packet to construct a sequence set; Calculate the difference between adjacent data points in the sequence set and input it into the classifier; If the classification decision value output by the classifier is less than negative one, then the corresponding difference value is determined to belong to the packet loss anomaly class; Count the number of values ​​that fall under the category of packet loss anomalies. If the ratio of this number to the total number of data points exceeds 30%, then a data packet transmission loss problem is confirmed. The classifier is a support vector machine, which is constructed using a radial basis function kernel.

9. A power distribution cabinet load data anomaly feedback verification system, characterized in that, The system includes: The data acquisition and time alignment module is used to acquire the local load measurement value and the remote feedback load data of the distribution cabinet, extract the timestamp to establish a time alignment mapping table, and perform linear interpolation compensation for mismatched data points to obtain time-synchronized load data pairs. The difference calculation and baseline extraction module is used to compare load data pairs point by point to calculate the numerical deviation, sort the difference sequence by time, divide the difference sequence by time period, and extract the maximum and average deviation of each time period as historical difference reference baseline. The anomaly detection and labeling module is used to determine the difference threshold based on the historical difference reference baseline, and to calculate the path length normalization coefficient of each value in the difference sequence using the isolated forest algorithm to obtain the anomaly score. If the anomaly score is greater than the preset threshold and the corresponding deviation value exceeds the preset multiple of the difference threshold, it is marked as a distortion anomaly marker point. The packet loss determination and time period recording module is used to extract the transmission delay within a preset time period before and after the distortion anomaly marker and calculate the fluctuation amplitude. If the fluctuation amplitude is greater than the preset fluctuation threshold and the peak delay exceeds the preset peak threshold, it is determined to be packet loss, and the start and end timestamps are recorded to form the abnormal time period. The data retransmission and difference update module is used to generate a data retransmission request based on the abnormal time period and send it to the remote device, receive the retransmission load data, extract the corresponding load value, re-insert it according to the original timestamp, and update the difference sequence. The resend verification and anomaly confirmation module is used to classify the updated difference sequences using a classifier, count the number of packet loss anomaly values, and if the proportion exceeds the preset proportion, it is confirmed that there is a data packet transmission loss problem. The alarm reporting and log recording module is used to extract the power distribution cabinet identification code and abnormal timestamp to form an alarm information data packet, which is sent to the monitoring center through the fault reporting channel to trigger equipment abnormality handling, and the packet loss event is recorded in the power distribution cabinet fault log database.

Images