A method and system for early warning of inspection risks of traffic vibration coupled power metering anomalies

Abstract

This invention relates to a method and system for early warning of inspection risks related to traffic vibration-coupled power metering anomalies, belonging to the field of power inspection by power supply companies. By constructing a voltage-current correlation baseline model, calculating the deviation index, and making horizontal comparisons, it identifies sub-healthy terminals at the critical point of increased contact resistance in advance. Traffic vibration sources are incorporated into a knowledge graph, establishing spatial correlation edges and calculating temporal correlation coefficients. Based on sub-health labels, spatial correlation, and temporal coupling characteristics, composite inference rules are constructed to determine the environmental coupling fault cause of current loss events, generating differentiated handling work orders accordingly, and updating the model confidence using feedback results. This solves the problem of misjudging accidents and electricity theft, improves the accuracy and adaptability of early warnings, significantly enhances the reliability of power inspection in complex environments, and overcomes the shortcomings of existing technologies in handling complex operating conditions, such as difficulty in distinguishing between accidents and electricity theft, frequent false alarms, and the inability to trace causal chains.

Description

Technical Field

[0001] This invention relates to a method and system for early warning of inspection risks of traffic vibration coupled power metering anomalies, belonging to the field of power inspection by power supply companies. Background Technology

[0002] In the current power inspection technology system, a long-standing but insufficiently addressed technical challenge lies in the frequent occurrence of metering anomalies and the difficulty in attributing the causes in areas along major transportation routes, including national highways, provincial highways, and expressways. In these areas, the terminals are subjected to continuous micro-vibrations from heavy vehicle traffic. Such mechanical stress disturbances easily lead to a gradual increase in terminal contact resistance, evolving from a healthy state to a sub-healthy state—that is, a critical increase in contact resistance without triggering a visible alarm—and ultimately developing into a failure state, manifested as current loss events. Existing technologies have several significant shortcomings in addressing this specific scenario: while traditional threshold alarm methods based on fixed rule bases have the advantage of clear logic, their rule bases are mostly statically set, only able to respond to visible faults exceeding preset thresholds. The system cannot effectively detect the gradual increase in contact resistance caused by the cumulative effect of traffic vibration. Such hidden risks remain outside the monitoring system for a long time before triggering explicit alarms, causing a large number of terminals in a sub-healthy state to be unable to be identified and intervened in a timely manner, thus creating potential faults. Although anomaly detection models based on data mining can extract statistical patterns from massive amounts of data, their black-box nature means they lack the ability to model the causal relationship between physical disturbances and electrical anomalies. When traffic vibration is superimposed on the deterioration of the equipment itself to form a compound working condition, such models are very likely to misjudge natural faults caused by vibration—such as current loss events caused by loose terminals—as electricity theft. This defect directly leads to a high false alarm rate in areas along major traffic arteries, resulting in a large number of invalid on-site inspections and a waste of human and material resources.

[0003] In the face of power outages caused by traffic vibrations, existing technologies cannot comprehensively utilize multi-dimensional evidence such as terminal sub-health status, spatial proximity, temporal correlation, and waveform characteristics for composite reasoning. This makes it difficult to scientifically distinguish between environmentally coupled faults—i.e., contact failures caused by traffic vibrations—and human-caused electricity theft, two completely different types of risks, at the level of risk attribution. This problem is particularly prominent in areas with frequent traffic vibrations, such as national and provincial highways. A large number of natural faults caused by environmental vibrations are misjudged as suspected electricity theft, which not only leads to a misallocation of inspection resources but may also induce power supply disputes and negatively impact user relationships. Summary of the Invention

[0004] Based on the problems described in the background, the present invention aims to solve the following problem: providing a method and system for early warning of inspection risks of traffic vibration coupled power metering anomalies. By sensing hidden risks, constructing environmental causal relationships and implementing closed-loop self-learning, the method solves the problem of misjudging accidents and electricity theft, improves the accuracy and adaptability of early warning, significantly enhances the reliability of power inspection in complex environments, and overcomes the shortcomings of existing technologies in handling complex operating conditions, such as difficulty in distinguishing between accidents and electricity theft, frequent false alarms, and inability to trace the causal chain.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for early warning of inspection risks of traffic vibration coupled power metering anomalies, comprising the following steps:

[0006] S1. Obtain power operation data and external environment data of the target metering point;

[0007] S2. Based on the preset historical health period data in the power operation data, construct a voltage-current correlation baseline model for each terminal; compare the real-time power operation data with the baseline model and calculate the voltage drop slope deviation index; based on the horizontal comparison of terminals of the same model in the same meter box, identify sub-healthy terminals that are in a critical state of increased contact resistance but have not triggered explicit alarms, and assign a first risk label to the sub-healthy terminals.

[0008] S3. Monitor and capture explicit electrical events in the power operation data; when an explicit electrical event is captured, record the event occurrence time and event type, and locate the associated target terminal.

[0009] S4. Materialize the traffic vibration sources in the external environment data and incorporate them into the knowledge graph. Calculate the spatial distance between each traffic vibration source and the location of the terminal block with the first risk label, and establish spatial association edges. Compare the occurrence time of the explicit electrical event with the peak traffic hours of heavy vehicles corresponding to the traffic vibration source in time series, and calculate the time correlation coefficient.

[0010] S5. Based on the spatial correlation edge, the temporal correlation coefficient, and the first risk label, construct a composite inference rule; based on the composite inference rule, determine the risk attribution of the current loss event in the explicit electrical event to obtain the risk attribution category;

[0011] S6. Generate differentiated handling work orders based on the risk attribution categories; feed back the execution results of the handling work orders to the knowledge graph, and update the status labels of the corresponding terminals and the confidence parameters of the composite inference rules.

[0012] Preferably, step S2 includes the following steps:

[0013] S2.1 Construct a baseline model of voltage-current correlation;

[0014] S2.2 Calculate the voltage drop slope deviation index;

[0015] S2.3 Identify sub-healthy individuals and assign them a primary risk label;

[0016] Step S2.1 includes the following steps:

[0017] Select the power operation data of the target metering point within a preset historical health period, where the preset historical health period is a period in which the terminal block operates stably and there are no abnormal records;

[0018] Power operation data within the preset historical health period is sampled to construct a voltage-current sample dataset for each terminal. ,in For the first Current values ​​at each sampling point For the first Voltage values ​​at each sampling point This represents the total number of sampling points;

[0019] Based on the aforementioned sample dataset, a baseline model of voltage-current correlation between the terminals is established using a linear regression method. This baseline model is expressed as follows: ,in For the model slope parameter, For model intercept parameters, This refers to the load current of the wiring terminals;

[0020] The residual sequence of each sampling point in the sample dataset relative to the baseline model is calculated using the following formula: ,in For the first Measured voltage values ​​at each sampling point;

[0021] Calculate the standard deviation of the residual series ;

[0022] The Mahalanobis distance of each sampling point is calculated using the following formula: ,in For the first The absolute value of the residuals at each sampling point;

[0023] The goodness-of-fit index of the model is determined by the following formula: ,in This represents the mean of the voltage sample data;

[0024] When the model fit index When the health status falls below the first preset threshold, the preset historical health period is reselected for modeling.

[0025] Preferably, step S2.2 includes the following steps:

[0026] Get the current sampling time Real-time current value With real-time voltage value And based on the voltage-current correlation baseline model of the terminals. Calculate the predicted voltage value at the current moment. , The current sampling time The real-time current value;

[0027] Calculate the residual at the current time step ;

[0028] Get continuous residual sequence at each sampling time point ,in The preset sliding time window length;

[0029] Calculate the mean of the residual sequence. Calculate the standard deviation of the residual sequence. ;

[0030] Calculate the voltage sag slope deviation index using the formula. ,in The absolute value of the mean of the residual sequence. The standard deviation of the residual sequence is denoted as . The standard deviation of the residual sequence obtained when constructing the baseline model is given. The preset volatility weighting coefficient;

[0031] When the voltage drop slope deviates from the exponential When the second preset threshold is exceeded, it is determined that the current real-time data deviates significantly from the baseline model.

[0032] Preferably, step S2.3 includes the following steps:

[0033] Identify and obtain the target terminal block, which is the sub-healthy terminal block that is in a critical state of increased contact resistance but has not triggered a visible alarm;

[0034] Based on the voltage drop slope deviation index of all terminals of the same model in the meter box, construct a set of deviation index datasets. ,in For the first The voltage drop slope deviation index of the same type of terminal block. This refers to the total number of terminals of the same type within the same meter box;

[0035] The mean of the deviation index dataset within the same group is calculated using the following formula: ; Calculate the standard deviation of the same group of deviation index datasets, using the following formula: ;

[0036] The lateral outlier coefficient of the target terminal block is calculated using the following formula: ,in The voltage drop slope deviation index of the target terminal;

[0037] When the lateral outlier coefficient When the value exceeds the third preset threshold, the target terminal is determined to be a sub-healthy terminal that is in a critical state of increased contact resistance but has not triggered a visible alarm.

[0038] Calculate the percentile ranking of the target terminal block in the same group of deviation index datasets. ;

[0039] When the percentile ranking The lateral outlier coefficient is greater than the fourth preset threshold and When the value exceeds the third preset threshold, a first risk label is assigned to the target terminal block.

[0040] when When the sample size is less than the fifth preset threshold, it is determined that the sample size of the terminal in the same group is insufficient. The longitudinal historical comparison is used instead of the horizontal comparison. The trend analysis is performed based on the historical deviation index sequence of the target terminal itself. When the voltage drop slope deviation index of multiple consecutive sampling periods shows a monotonically increasing trend and the current value exceeds the sixth preset threshold by a multiple of the historical average, the target terminal is given a first risk label.

[0041] Preferably, step S3 includes the following steps:

[0042] The explicit electrical events include voltage sag / boost events, current loss events, and harmonic surge events;

[0043] Record the occurrence time of the aforementioned explicit electrical events. and event types And extract the occurrence time of the explicit electrical event. Electrical characteristic data within the preset time window before and after;

[0044] The event impact phase is determined based on the electrical characteristic data. ;

[0045] When one of the following conditions is met: the phase voltage sag exceeds the seventh preset threshold, the phase current suddenly drops to near zero, or the phase harmonic content increases before and after the event exceeds the eighth preset threshold, the phase is determined to be the phase affected by the event.

[0046] Based on the phase affected by the event and the time of the event, electrical topology information pre-stored in the knowledge graph is obtained and combined with the knowledge graph. A set of candidates with direct electrical connection with the phase affected by the event is retrieved from the knowledge graph.

[0047] Obtain the time of occurrence of each candidate terminal in the candidate terminal block set during the dominant electrical event. The electrical characteristic data within the preset time window before and after the test are used to calculate the characteristic change of each candidate terminal. The calculation formula is: ,in The first time window before the event occurs The average electrical characteristic parameters of each candidate terminal;

[0048] Within a preset time window after the event occurs The average electrical characteristic parameters of each candidate terminal;

[0049] According to the formula Calculate the correlation index. Candidate terminals exceeding the ninth preset threshold are identified as suspected associated terminals;

[0050] Obtain the geographical location and electrical path distance of the suspected associated terminals. According to the formula Calculate the overall association confidence score, where For preset weighting coefficients, The maximum path distance. As an indicator variable;

[0051] The terminal with the highest overall association confidence is identified as the terminal associated with the explicit electrical event, and it is associated with the event information and stored in the knowledge graph.

[0052] Preferably, step S4 includes the following steps:

[0053] S4.1 Establish spatially related edges;

[0054] S4.2 Calculate the time correlation coefficient;

[0055] Step S4.1 includes the following steps:

[0056] Obtain traffic vibration source information from the external environment data, and incorporate each traffic vibration source as an independent entity into the knowledge graph;

[0057] Obtain the geographic coordinates of each terminal block bearing the first risk label. ;

[0058] Calculate the first according to the formula. The traffic vibration source and the first Spacing between terminals bearing the first risk label ,in For the Earth's radius, and The first The latitude and longitude of each terminal block. and The first The latitude and longitude of each traffic vibration source;

[0059] When the spatial distance When it is less than the tenth preset threshold, determine the first The terminal block is located at the first Within the influence range of a traffic vibration source;

[0060] Calculate the first according to the formula. Comprehensive vibration impact index of each terminal block ,in The preset weighting coefficients and , The preset maximum influence distance, For the first Normalized value of the proportion of heavy vehicles among traffic vibration sources For the first Normalized values ​​of vibration intensity levels of individual traffic vibration sources;

[0061] When the spatial distance When the value is less than the tenth preset threshold, it is the first in the knowledge graph. The traffic vibration source entity and the first Establish spatial association edges between the terminal block entities;

[0062] Set attribute fields for the spatially associated edges; store the spatially associated edges and their attribute information in the knowledge graph, and update the association information field of the terminal block entity, adding the list of traffic vibration sources associated with it and the corresponding comprehensive vibration impact index.

[0063] Preferably, step S4.2 includes the following steps:

[0064] Obtain traffic vibration source entities that have spatially associated edges with the target terminal block, and extract the set of peak traffic periods for heavy vehicles associated with the traffic vibration sources. Each peak period is defined by its start time. and end time definition;

[0065] Obtain the occurrence time of the explicit electrical event. And convert it to the same time scale as the peak period;

[0066] Calculate the first The time overlap between the peak period and the time of the event. The time correlation coefficient is calculated using the following formula: The calculated time correlation coefficient As attribute information, it is stored in the spatial association edge between the target terminal block and the traffic vibration source in the knowledge graph.

[0067] Preferably, the determination conditions for the compound reasoning rule in step S5 include:

[0068] The first condition is that the target terminal corresponding to the current loss event has the first risk label;

[0069] The second condition is that the target terminal block and the traffic vibration source have a spatially related edge.

[0070] The third condition is the temporal correlation coefficient between the occurrence time of the flow loss event and the peak period of heavy vehicle traffic at the traffic vibration source. Greater than the eleventh preset threshold;

[0071] The fourth condition is that, after analyzing the waveform characteristics of the current loss event, it is confirmed that it is accompanied by a sudden increase in harmonic content or voltage flicker.

[0072] When the first, second, third, and fourth conditions are met simultaneously, the risk attribution category of the current loss event is determined to be an environmental coupling failure; the determination result of the composite reasoning rule is used as the basis for updating the status label of the target terminal.

[0073] Preferably, step S6 includes the following steps:

[0074] When the risk attribution category of the power loss event is determined to be an environmental coupling failure based on the composite reasoning rule, an equipment maintenance work order is generated.

[0075] An inspection work order is generated when the risk attribution category of the current loss event does not meet all the judgment conditions of environmental coupling failure and there are no other clear attributions, or when the characteristics of electricity theft are detected.

[0076] Obtain the on-site verification results fed back after the completion of the processing work order; update the status label of the target terminal in the knowledge graph based on the on-site verification results;

[0077] Based on the comparison between the on-site verification results and the reasoning conclusions, update the confidence parameter of the composite reasoning rule;

[0078] The updated confidence parameters are stored as attributes of the composite inference rule in the knowledge graph.

[0079] This invention also provides a traffic vibration coupled type power metering anomaly inspection risk early warning system, comprising:

[0080] The data acquisition module obtains power operation data and external environmental data at the target metering point;

[0081] The terminal status diagnosis module constructs a voltage-current correlation baseline model for each terminal based on the preset historical health period data in the power operation data; compares the real-time power operation data with the baseline model to calculate the voltage drop slope deviation index; and identifies sub-healthy terminals that are in a critical state of increased contact resistance but have not triggered a visible alarm based on a horizontal comparison of terminals of the same model in the same meter box, and assigns a first risk label to the sub-healthy terminals.

[0082] The electrical event monitoring and location module monitors and captures visible electrical events in the power operation data; when a visible electrical event is captured, it records the event occurrence time and event type, and locates the associated target terminal.

[0083] The environmental correlation analysis module materializes traffic vibration sources in external environmental data and incorporates them into a knowledge graph. It calculates the spatial distance between each traffic vibration source and the location of the terminal block with the first risk label, and establishes spatial correlation edges. It also performs a time-series comparison between the occurrence time of the explicit electrical event and the peak traffic hours of the corresponding traffic vibration source for heavy vehicles, and calculates the time correlation coefficient.

[0084] The composite reasoning decision module constructs composite reasoning rules based on the spatial correlation edge, the temporal correlation coefficient, and the first risk label; and performs risk attribution determination on the current loss event in the explicit electrical event based on the composite reasoning rules to obtain the risk attribution category.

[0085] The work order processing and feedback learning module generates differentiated work orders based on the risk attribution category; it feeds back the execution results of the work orders to the knowledge graph and updates the status labels of the corresponding terminals and the confidence parameters of the composite inference rules.

[0086] The beneficial effects of this invention are:

[0087] 1. This invention constructs a voltage-current correlation baseline model for each terminal based on preset historical health period data, and identifies sub-healthy terminals in a critical state of increasing contact resistance based on lateral comparison of terminals of the same model in the same meter box, thereby achieving refined online perception of the hidden deterioration state of equipment. This micro-state perception mechanism enables a large number of "sub-healthy" terminals that have been outside the monitoring field of vision for a long time to be assigned the first risk label, effectively overcoming the problem of early warning lag caused by traditional methods that rely only on explicit alarms and cannot perceive progressive hidden risks.

[0088] 2. This invention materializes traffic vibration sources in external environmental data and incorporates them into a knowledge graph, establishing a spatial association edge between the traffic vibration source and the terminal block with the first risk label. It also calculates the time correlation coefficient by comparing the occurrence time of the explicit electrical event with the peak traffic period of heavy vehicles, thus realizing the causal relationship modeling between physical disturbances and electrical anomalies. This composite reasoning mechanism based on both spatial and temporal associations can make a comprehensive judgment based on the first risk label, spatial association edge, time correlation coefficient, and waveform characteristics when capturing power loss events. It can scientifically distinguish between environmental coupling faults and human power theft, two completely different types of risk attribution, and solve the problem of frequent false alarms and inability to trace the causal chain in the existing technology under composite working conditions.

[0089] 3. This invention feeds back the execution results of the handling work order to the knowledge graph, updates the status labels of the corresponding terminals and the confidence parameters of the composite reasoning rules, and realizes a closed-loop optimization mechanism that continuously learns and evolves from each on-site inspection. When the on-site inspection results are consistent with the reasoning conclusion, the confidence is increased; when they are inconsistent, the confidence is decreased. This allows the composite reasoning rules to be continuously corrected and optimized with the accumulation of data, significantly improving the system's early warning accuracy, decision interpretability, and adaptability in the identification of metering anomalies in areas with frequent traffic vibrations, such as national highways and provincial highways. Attached Figure Description

[0090] Figure 1 This is a flowchart of the method steps of the present invention;

[0091] Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0092] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0093] Example 1

[0094] like Figure 1 As shown, this invention provides a method for early warning of inspection risks of traffic vibration coupled power metering anomalies, comprising the following steps:

[0095] S1: Obtain power operation data and external environment data of the target metering point;

[0096] S2: Based on the preset historical health period data in the power operation data, construct a voltage-current correlation baseline model for each terminal; compare the real-time power operation data with the baseline model and calculate the voltage drop slope deviation index; based on the horizontal comparison of terminals of the same model in the same meter box, identify sub-healthy terminals that are in a critical state of increased contact resistance but have not triggered explicit alarms, and assign a first risk label to the sub-healthy terminals.

[0097] S3: Monitor and capture explicit electrical events in the power operation data; when an explicit electrical event is captured, record the event occurrence time and event type, and locate the associated target terminal;

[0098] S4: Materialize the traffic vibration sources in the external environment data and incorporate them into the knowledge graph; calculate the spatial distance between each traffic vibration source and the location of the terminal block with the first risk label; establish spatial association edges; compare the occurrence time of the explicit electrical event with the peak traffic hours of heavy vehicles corresponding to the traffic vibration source in time series and calculate the time correlation coefficient.

[0099] S5: Based on the spatial correlation edge, the temporal correlation coefficient, and the first risk label, construct a composite inference rule; based on the composite inference rule, determine the risk attribution of the current loss event in the explicit electrical event to obtain the risk attribution category;

[0100] S6: Generate differentiated handling work orders based on the risk attribution categories; feed back the execution results of the handling work orders to the knowledge graph, and update the status labels of the corresponding terminals and the confidence parameters of the composite inference rules.

[0101] Step S2 includes the following steps:

[0102] S2.1 Construct a baseline model of voltage-current correlation;

[0103] S2.2 Calculate the voltage drop slope deviation index;

[0104] S2.3 Identify sub-healthy individuals and assign them a primary risk label;

[0105] Step S2.1 includes the following steps:

[0106] Select the power operation data of the target metering point within a preset historical health period, where the preset historical health period is a period in which the terminal block operates stably and there are no abnormal records;

[0107] Power operation data within the preset historical health period is sampled to construct a voltage-current sample dataset for each terminal. ,in For the first Current values ​​at each sampling point For the first Voltage values ​​at each sampling point This represents the total number of sampling points;

[0108] Based on the aforementioned sample dataset, a baseline model of voltage-current correlation between the terminals is established using a linear regression method. This baseline model is expressed as follows: ,in For the model slope parameter, For model intercept parameters, This refers to the load current of the wiring terminals;

[0109] The residual sequence of each sampling point in the sample dataset relative to the baseline model is calculated using the following formula: ,in For the first Measured voltage values ​​at each sampling point;

[0110] Calculate the standard deviation of the residual series ;

[0111] The Mahalanobis distance of each sampling point is calculated using the following formula: ,in For the first The absolute value of the residuals at each sampling point;

[0112] The goodness-of-fit index of the model is determined by the following formula: ,in This represents the mean of the voltage sample data;

[0113] When the model fit index When the health status falls below the first preset threshold, the preset historical health period is reselected for modeling.

[0114] Regarding step S2.1, it should be explained that the target metering point is a specific electricity metering device installed on the electricity user side that requires risk monitoring, and its unique identifier is obtained from the equipment ledger database of the marketing business application system.

[0115] The preset historical health period is determined by querying historical operation and maintenance records and alarm information to determine the time period. The operation data of the terminal for the past 6 months is retrieved from the power consumption information collection system. Dates with alarms or troubleshooting records that show abnormal voltage, current imbalance, or terminal overheating are removed from the inspection records. A continuous time interval without any abnormal events and with a relatively stable load curve is selected as the health period.

[0116] With current value The independent variable is the voltage value. Using the least squares method as the dependent variable, a univariate linear regression is performed, which mathematically boils down to finding a straight line. This makes all sample points The sum of squared distances to this line is minimized; slope parameter and intercept parameter Using linear regression, based on the sample dataset The optimal fitting parameters are automatically calculated; where Characterizes the voltage drop caused by a unit change in current under healthy conditions. The reference voltage characterizing the no-load condition;

[0117] The formula for calculating the standard deviation of a residual sequence is: ,in Let be the mean of the residual sequence. Represents the square root operation; For the first The residual at each sampling point represents the deviation between the measured voltage value at that sampling point and the predicted voltage value calculated using the baseline model. represents the degrees of freedom of the sample standard deviation, where This represents the total number of sampling points;

[0118] The Mahalanobis distance of each sampling point is used to evaluate the degree of deviation of each sampling point from the baseline model; the first preset threshold is obtained by collecting historical operating data of the same type of terminal block in the local power grid that is in a healthy operating state, constructing the baseline model respectively and calculating their respective goodness-of-fit indices; statistically analyzing the distribution of the goodness-of-fit indices of healthy terminals, and taking the lower limit of the index covering more than 95% of healthy samples according to industry standards as the first preset threshold.

[0119] When the model fit index If the data falls below a first preset threshold, it is determined that the data of the preset historical health period does not meet the modeling requirements, and the preset historical health period is reselected for modeling.

[0120] Step S2.2 includes the following steps:

[0121] Get the current sampling time Real-time current value With real-time voltage value And based on the voltage-current correlation baseline model of the terminals. Calculate the predicted voltage value at the current moment. , The current sampling time The real-time current value;

[0122] Calculate the residual at the current time step ;

[0123] Get continuous residual sequence at each sampling time point ,in The preset sliding time window length;

[0124] Calculate the mean of the residual sequence. Calculate the standard deviation of the residual sequence. ;

[0125] Calculate the voltage sag slope deviation index using the formula. ,in The absolute value of the mean of the residual sequence. The standard deviation of the residual sequence is denoted as . The standard deviation of the residual sequence obtained when constructing the baseline model is given. The preset volatility weighting coefficient;

[0126] When the voltage drop slope deviates from the exponential When the second preset threshold is exceeded, it is determined that the current real-time data deviates significantly from the baseline model.

[0127] Regarding step S2.2, it needs to be explained that the... The input variables for calculating the current voltage prediction value are substituted into the voltage-current correlation baseline model; the formula for calculating the residual at the current moment is: ,in The real-time voltage value, The voltage prediction value at the current moment is given; the preset sliding time window length M is used to determine the number of continuous sampling points participating in the real-time deviation index calculation. The acquisition method is based on the typical sampling frequency of the electricity information acquisition system, such as collecting one data point per minute, and combined with the time span that can effectively capture voltage fluctuation trends without being sensitive in engineering experience, such as 10 minutes, and M is set to 10.

[0128] The formula for calculating the mean of the residual sequence is: ,in For the continuous The th sampling time residual sequence in the th sampling time residual sequence The nth residual represents the nth residual. The deviation between the measured voltage value at each sampling time and the predicted voltage value calculated using the baseline model. The formula for calculating the standard deviation of the residual sequence is: ,in The degrees of freedom for the sample standard deviation. The length of the sliding time window;

[0129] In the formula for calculating the standard deviation Representing the The deviation between the residual value at each sampling point and the average residual value within the current sliding time window reflects the degree of deviation between the measured voltage value and the baseline model prediction value at the sampling time, relative to the recent average deviation level; in the voltage drop slope deviation exponent formula... Characterizes the overall direction and extent of the shift in real-time data relative to the baseline model. Characterizing the volatility of real-time data relative to the baseline model, Used to normalize the deviation index to the volatility level of historical healthy periods. To balance the contributions of deviation degree and volatility to the deviation index, terminal samples labeled with healthy and sub-healthy states are obtained from a historical database. Using the sample labels as the target variable, a logistic regression model is used for binary classification training. The feature weight ratio output by the model after training is the preset volatility weight coefficient. ;

[0130] The second preset threshold is obtained by acquiring terminal block samples marked as normal and significantly deviating from the historical database, using the sample labels as the target variable, and training a decision tree classification model. After training, the classification threshold of the root node of the model is the second preset threshold.

[0131] Step S2.3 includes the following steps:

[0132] Identify and obtain the target terminal block, which is the sub-healthy terminal block that is in a critical state of increased contact resistance but has not triggered a visible alarm;

[0133] Based on the voltage drop slope deviation index of all terminals of the same model in the meter box, construct a set of deviation index datasets. ,in For the first The voltage drop slope deviation index of the same type of terminal block. This refers to the total number of terminals of the same type within the same meter box;

[0134] The mean of the deviation index dataset within the same group is calculated using the following formula: ; Calculate the standard deviation of the same group of deviation index datasets, using the following formula: ;

[0135] The lateral outlier coefficient of the target terminal block is calculated using the following formula: ,in The voltage drop slope deviation index of the target terminal;

[0136] When the lateral outlier coefficient When the value exceeds the third preset threshold, the target terminal is determined to be a sub-healthy terminal that is in a critical state of increased contact resistance but has not triggered a visible alarm.

[0137] Calculate the percentile ranking of the target terminal block in the same group of deviation index datasets. ;

[0138] When the percentile ranking The lateral outlier coefficient is greater than the fourth preset threshold and When the value exceeds the third preset threshold, a first risk label is assigned to the target terminal block.

[0139] when When the sample size is less than the fifth preset threshold, it is determined that the sample size of the terminal in the same group is insufficient. The longitudinal historical comparison is used instead of the horizontal comparison. The trend analysis is performed based on the historical deviation index sequence of the target terminal itself. When the voltage drop slope deviation index of multiple consecutive sampling periods shows a monotonically increasing trend and the current value exceeds the sixth preset threshold by a multiple of the historical average, the target terminal is given a first risk label.

[0140] Regarding step S2.3, it needs to be explained that the target terminal block involves periodically traversing all terminal block entities stored in the knowledge graph, and using the currently traversed terminal block as the target terminal block; the formula for calculating the mean of the same group deviation index dataset is: The standard deviation of the same group of deviation index datasets is calculated using the following formula: The lateral outlier coefficient characterizes the degree of deviation of the target terminal relative to its position within the same group of terminals.

[0141] The third preset threshold is determined by obtaining terminal block samples labeled with sub-health and normal states from a historical database, with each sample's feature being the lateral outlier coefficient; using the sample label as the target variable, a decision tree classification model is trained, and the classification threshold of the model's root node after training is the third preset threshold; and the percentile ranking of the target terminal block in the same group's deviation index dataset. The calculation formula is: ,in The voltage drop slope deviation index of the target terminal is in the same deviation index dataset. The positions in ascending order;

[0142] The fourth preset threshold involves retrieving terminal block samples marked as being in sub-healthy or normal states from a historical database, with each sample characterized by its percentile ranking. Using sample labels as the target variable, a decision tree classification model is used for training. After training, the classification threshold of the root node of the model is the fourth preset threshold.

[0143] The first risk label includes the lateral outlier coefficient. and the percentile ranking The fifth preset threshold is obtained from the historical database. For terminal samples within the same group under the same value, the consistency index of the results of sub-health determination of the same batch of terminals was calculated based on both horizontal and vertical comparison methods; The values ​​are independent variables, and the consistency index is the dependent variable. A linear regression model is used for fitting, and the value corresponding to the first time the consistency index on the fitted curve reaches a high probability, such as 0.9, is selected. The value is used as the fifth preset threshold;

[0144] The sixth preset threshold is obtained from the historical database of terminal samples that have been marked as sub-healthy or normal and whose number of terminals in the same group is less than the fifth preset threshold. The feature of each sample is the multiple by which the current value exceeds the historical average. Using the sample label as the target variable, a logistic regression model is used for binary classification training. The multiple corresponding to the model output probability of 0.5 after training is the sixth preset threshold.

[0145] Step S3 includes the following steps:

[0146] The explicit electrical events include voltage sag / boost events, current loss events, and harmonic surge events;

[0147] Record the occurrence time of the aforementioned explicit electrical events. and event types And extract the occurrence time of the explicit electrical event. Electrical characteristic data within the preset time window before and after;

[0148] The event impact phase is determined based on the electrical characteristic data. ;

[0149] When one of the following conditions is met: the phase voltage sag exceeds the seventh preset threshold, the phase current suddenly drops to near zero, or the phase harmonic content increases before and after the event exceeds the eighth preset threshold, the phase is determined to be the phase affected by the event.

[0150] Based on the phase affected by the event and the time of the event, electrical topology information pre-stored in the knowledge graph is obtained and combined with the knowledge graph. A set of candidates with direct electrical connection with the phase affected by the event is retrieved from the knowledge graph.

[0151] Obtain the time of occurrence of each candidate terminal in the candidate terminal block set during the dominant electrical event. The electrical characteristic data within the preset time window before and after the test are used to calculate the characteristic change of each candidate terminal. The calculation formula is: ,in The first time window before the event occurs The average electrical characteristic parameters of each candidate terminal;

[0152] Within a preset time window after the event occurs The average electrical characteristic parameters of each candidate terminal;

[0153] According to the formula Calculate the correlation index. Candidate terminals exceeding the ninth preset threshold are identified as suspected associated terminals;

[0154] Obtain the geographical location and electrical path distance of the suspected associated terminals. According to the formula Calculate the overall association confidence score, where For preset weighting coefficients, The maximum path distance. As an indicator variable;

[0155] The terminal with the highest overall association confidence is identified as the terminal associated with the explicit electrical event, and it is associated with the event information and stored in the knowledge graph.

[0156] Regarding step S3, it needs to be explained that when abnormal features in the power operation data that meet the judgment rules for voltage sag / surge events, current loss events, or harmonic surge events are detected, a significant electrical event is determined to be captured. The pre- and post-preset time windows are obtained from the historical database by acquiring samples of significant electrical events whose affected phases have been marked. The electrical feature data extracted under different time window lengths are used as input features, and the correctness of the determination of the affected phases is used as the target variable. A decision tree classification model is used for training. After training, the optimal time window length selected by the model is the pre- and post-preset time window.

[0157] The seventh preset threshold is obtained from the historical database by acquiring phase samples of the marked event-affected phase and the non-event-affected phase, with each sample characterized by the phase voltage sag amplitude; the sample label is used as the target variable, and a decision tree classification model is used for training. After training, the classification threshold of the root node of the model is the seventh preset threshold.

[0158] The eighth preset threshold is obtained from the historical database, which contains phase samples of the marked event-affected phase and the non-event-affected phase. The feature of each sample is the incremental value of the harmonic content before and after the event. The sample label is used as the target variable, and a decision tree classification model is used for training. After training, the classification threshold of the root node of the model is the eighth preset threshold. The electrical topology information includes the electrical connection relationship between metering points, meter boxes, and wiring terminals.

[0159] The ninth preset threshold is used to obtain candidate terminal samples marked as associated or unassociated from the historical database. The feature of each sample is an association degree index. Using sample labels as the target variable, a decision tree classification model is used for training. After training, the classification threshold of the root node of the model is the ninth preset threshold.

[0160] The Preset weighting coefficients are used to retrieve suspected related terminal samples marked as related or unrelated from the historical database. The feature vector of each sample is generated by... , and Composition; using sample labels as target variables, a logistic regression model is used for binary classification training. After training, the weight coefficients of the three features output by the model are normalized to obtain the composition. Preset weighting coefficients; The value is 1 when the terminal has the first risk label, and 0 when it does not have the first risk label.

[0161] Step S4 includes the following steps:

[0162] S4.1 Establish spatially related edges;

[0163] S4.2 Calculate the time correlation coefficient;

[0164] Step S4.1 includes the following steps:

[0165] Obtain traffic vibration source information from the external environment data, and incorporate each traffic vibration source as an independent entity into the knowledge graph;

[0166] Obtain the geographic coordinates of each terminal block bearing the first risk label. ;

[0167] Calculate the first according to the formula. The traffic vibration source and the first Spacing between terminals bearing the first risk label ,in For the Earth's radius, and The first The latitude and longitude of each terminal block. and The first The latitude and longitude of each traffic vibration source;

[0168] When the spatial distance When it is less than the tenth preset threshold, determine the first The terminal block is located at the first Within the influence range of a traffic vibration source;

[0169] Calculate the first according to the formula. Comprehensive vibration impact index of each terminal block ,in The preset weighting coefficients and , The preset maximum influence distance, For the first Normalized value of the proportion of heavy vehicles among traffic vibration sources For the first Normalized values ​​of vibration intensity levels of individual traffic vibration sources;

[0170] When the spatial distance When the value is less than the tenth preset threshold, it is the first in the knowledge graph. The traffic vibration source entity and the first Establish spatial association edges between the terminal block entities;

[0171] Set attribute fields for the spatially associated edges; store the spatially associated edges and their attribute information in the knowledge graph, and update the association information field of the terminal block entity, adding the list of traffic vibration sources associated with it and the corresponding comprehensive vibration impact index.

[0172] Regarding step S4.1, it should be explained that an attribute field is set for each traffic vibration source entity, and the attribute field includes a unique identifier for the vibration source. Geographical coordinates The geographical coordinates are determined based on the meter box installation location information in the equipment ledger data, including road type, traffic volume level, proportion of heavy vehicles, peak hours, and vibration intensity level.

[0173] The tenth preset threshold retrieves terminal samples marked as affected or unaffected by vibration from a historical database. Each sample is characterized by the spatial distance between the terminal and the nearest traffic vibration source. Using sample labels as the target variable, a decision tree classification model is used for training. After training, the classification threshold of the root node of the model is the tenth preset threshold.

[0174] The Preset weighting coefficients are used to obtain terminal block samples from the historical database that have been marked as having experienced environmental coupling faults or not, and are located within the influence range of traffic vibration sources. The feature vector of each sample is generated by... , and Composition; using sample labels as target variables, a logistic regression model is used for binary classification training. After training, the weight coefficients of the three features output by the model are normalized so that their sum is 1, which is the... Preset weighting coefficients;

[0175] The spatial association edge is a bidirectional edge, used to characterize the mutual influence between the traffic vibration source and the terminal block; the attribute field includes spatial distance. Comprehensive vibration impact index Establish timestamps and influence relationship types; for terminals simultaneously located within the influence range of multiple traffic vibration sources, establish multiple spatial association edges, and base them on the comprehensive vibration influence index. Prioritize each associated edge; the influence relationship type is set to be affected by vibration, which is used to qualitatively describe the association between the traffic vibration source and the terminal block.

[0176] Step S4.2 includes the following steps:

[0177] Obtain traffic vibration source entities that have spatially associated edges with the target terminal block, and extract the set of peak traffic periods for heavy vehicles associated with the traffic vibration sources. Each peak period is defined by its start time. and end time definition;

[0178] Obtain the occurrence time of the explicit electrical event. And convert it to the same time scale as the peak period;

[0179] Calculate the first The time overlap between the peak period and the time of the event. The time correlation coefficient is calculated using the following formula: The calculated time correlation coefficient As attribute information, it is stored in the spatial association edge between the target terminal block and the traffic vibration source in the knowledge graph.

[0180] Regarding step S4.2, it needs to be explained that the preset event impact time window half-width... To define the time range of an event affected by vibration, a decision tree model was trained. This model retrieved labeled samples of visible electrical events, both vibration-related and vibration-independent, from a historical database. The time correlation coefficients calculated under different time window half-widths were used as input features, with the sample labels as the target variable. The model was trained using a decision tree classification model, and the optimal time window half-width selected after training was determined. The aforementioned peak traffic periods for heavy vehicles Each peak period The peak period information is directly extracted from the attribute fields of the traffic vibration source entity, and the information comes from the road traffic monitoring data provided by the traffic management department; the occurrence time of the explicit electrical event is directly obtained from the event timestamp recorded by the event monitoring module.

[0181] The time overlap The calculation formula is: ,molecular Indicates the time window of event impact and the first The length of the overlapping intervals during peak periods, with a negative value of 0 indicating no overlap; the denominator... The total length of the event impact time window is used to normalize the overlap to the [0, 1] interval; the time correlation coefficient In the calculation formula, the molecule The sum of the overlap between all peak periods and event time windows, denominator This is the total number during peak hours; the final result is obtained from this ratio. value; The closer the value is to 1, the higher the match between the event's time and the peak traffic period for heavy vehicles; the closer it is to 0, the lower the match. The calculated... The value is stored as attribute information in the corresponding spatial association edge in the knowledge graph, and is used for the determination of the third condition in subsequent compound reasoning rules;

[0182] The determination conditions for the compound reasoning rule in step S5 include:

[0183] The first condition is that the target terminal corresponding to the current loss event has the first risk label;

[0184] The second condition is that the target terminal block and the traffic vibration source have a spatially related edge.

[0185] The third condition is the temporal correlation coefficient between the occurrence time of the flow loss event and the peak period of heavy vehicle traffic at the traffic vibration source. Greater than the eleventh preset threshold;

[0186] The fourth condition is that, after analyzing the waveform characteristics of the current loss event, it is confirmed that it is accompanied by a sudden increase in harmonic content or voltage flicker.

[0187] When the first, second, third, and fourth conditions are met simultaneously, the risk attribution category of the current loss event is determined to be an environmental coupling failure; the determination result of the composite reasoning rule is used as the basis for updating the status label of the target terminal.

[0188] Regarding step S5, it should be explained that the eleventh preset threshold is obtained through training a decision tree model. It retrieves labeled samples of current loss events (both environmentally coupled and non-environmentally coupled) from a historical database, with each sample's feature being a time correlation coefficient. Using sample labels as the target variable, a decision tree classification model is used for training. After training, the classification threshold of the root node of the model is the eleventh preset threshold.

[0189] The two characteristics in the fourth condition are as follows: the sudden increase in harmonic content is identified by the Fast Fourier Transform algorithm, which performs spectral analysis on the voltage and current waveforms for 10 cycles before and after the event to calculate the total harmonic distortion rate and the content of the 3rd, 5th, and 7th characteristic harmonics. When the harmonic content after the event increases by more than 20% compared to before the event, it is determined to be a sudden increase in harmonic content. The voltage flicker phenomenon is identified by detecting the fluctuation envelope of the effective voltage value. The voltage waveform is demodulated to obtain an amplitude-modulated wave. When the amplitude of the amplitude-modulated wave changes at a frequency within the range of 1-10Hz and the fluctuation amplitude exceeds 3% of the nominal voltage, it is determined that voltage flicker exists. Both characteristics are directly output by existing power quality analysis devices.

[0190] Step S6 includes the following steps:

[0191] When the risk attribution category of the power loss event is determined to be an environmental coupling failure based on the composite reasoning rule, an equipment maintenance work order is generated.

[0192] An inspection work order is generated when the risk attribution category of the current loss event does not meet all the judgment conditions of environmental coupling failure and there are no other clear attributions, or when the characteristics of electricity theft are detected.

[0193] Obtain the on-site verification results fed back after the completion of the processing work order; update the status label of the target terminal in the knowledge graph based on the on-site verification results;

[0194] Based on the comparison between the on-site verification results and the reasoning conclusions, update the confidence parameter of the composite reasoning rule;

[0195] The updated confidence parameters are stored as attributes of the composite inference rule in the knowledge graph.

[0196] Regarding step S6, it should be explained that the equipment maintenance work order includes a unique identifier for the target terminal block, a fault type label, causal chain explanation information, and on-site maintenance operation instructions.

[0197] The causal chain explanation information is generated according to the composite inference rule, including the identification criteria for the target terminal bearing the first risk label, the associated traffic vibration source information, and the time correlation coefficient. Numerical and waveform characteristic analysis conclusions;

[0198] The inspection work order includes the unique identifier of the target terminal, the suspected type of electricity theft, the period of abnormal data, and the key points of on-site verification; the on-site verification results include fault confirmation information, maintenance records, electricity theft confirmation status, and handling conclusions; if it is confirmed to be an environmental coupling fault, a historical fault tag is added to the terminal and the fault cause and maintenance time are recorded; if it is confirmed to be an electricity theft, an electricity theft record tag is added to the terminal and the investigation time is recorded;

[0199] When the on-site verification results are consistent with the reasoning conclusion, according to the formula... Increase confidence, where The current confidence level of the rule. This is the preset learning step size; when the on-site verification results are inconsistent with the reasoning conclusion, the formula is used... Lower the confidence level.

[0200] Example 2: Based on Example 1, the present invention also provides a traffic vibration coupled type power metering anomaly inspection risk early warning system, such as... Figure 2 As shown, it includes:

[0201] The data acquisition module obtains power operation data and external environmental data at the target metering point;

[0202] The terminal status diagnosis module constructs a voltage-current correlation baseline model for each terminal based on the preset historical health period data in the power operation data; compares the real-time power operation data with the baseline model to calculate the voltage drop slope deviation index; and identifies sub-healthy terminals that are in a critical state of increased contact resistance but have not triggered a visible alarm based on a horizontal comparison of terminals of the same model in the same meter box, and assigns a first risk label to the sub-healthy terminals.

[0203] The electrical event monitoring and location module monitors and captures visible electrical events in the power operation data; when a visible electrical event is captured, it records the event occurrence time and event type, and locates the associated target terminal.

[0204] The environmental correlation analysis module materializes traffic vibration sources in external environmental data and incorporates them into a knowledge graph. It calculates the spatial distance between each traffic vibration source and the location of the terminal block with the first risk label, and establishes spatial correlation edges. It also performs a time-series comparison between the occurrence time of the explicit electrical event and the peak traffic hours of the corresponding traffic vibration source for heavy vehicles, and calculates the time correlation coefficient.

[0205] The composite reasoning decision module constructs composite reasoning rules based on the spatial correlation edge, the temporal correlation coefficient, and the first risk label; and performs risk attribution determination on the current loss event in the explicit electrical event based on the composite reasoning rules to obtain the risk attribution category.

[0206] The work order processing and feedback learning module generates differentiated work orders based on the risk attribution category; it feeds back the execution results of the work orders to the knowledge graph and updates the status labels of the corresponding terminals and the confidence parameters of the composite inference rules.

Claims

1. A method for early warning of inspection risks of traffic vibration coupled-type power metering anomalies, characterized in that, Includes the following steps: S1. Obtain power operation data and external environment data of the target metering point; S2. Based on the preset historical health period data in the power operation data, construct a voltage-current correlation baseline model for each terminal; compare the real-time power operation data with the baseline model and calculate the voltage drop slope deviation index; based on the horizontal comparison of terminals of the same model in the same meter box, identify sub-healthy terminals that are in a critical state of increased contact resistance but have not triggered explicit alarms, and assign a first risk label to the sub-healthy terminals. S3. Monitor and capture explicit electrical events in the power operation data; When the explicit electrical event is captured, the event occurrence time and event type are recorded, and the associated target terminal is located; S4. Materialize the traffic vibration sources in the external environment data and incorporate them into the knowledge graph. Calculate the spatial distance between each traffic vibration source and the location of the terminal block with the first risk label, and establish spatial association edges. Compare the occurrence time of the explicit electrical event with the peak traffic hours of heavy vehicles corresponding to the traffic vibration source in time series, and calculate the time correlation coefficient. S5. Based on the spatial correlation edge, the temporal correlation coefficient, and the first risk label, construct a composite inference rule; based on the composite inference rule, determine the risk attribution of the current loss event in the explicit electrical event to obtain the risk attribution category; S6. Generate differentiated handling work orders based on the risk attribution categories; The execution result of the processing work order is fed back to the knowledge graph to update the status label of the corresponding terminal block and the confidence parameter of the composite inference rule.

2. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 1, characterized in that, Step S2 includes the following steps: S2.1 Construct a baseline model of voltage-current correlation; S2.2 Calculate the voltage drop slope deviation index; S2.3 Identify sub-healthy individuals and assign them a primary risk label; Step S2.1 includes the following steps: Select the power operation data of the target metering point within a preset historical health period, where the preset historical health period is a period in which the terminal block operates stably and there are no abnormal records; Power operation data within the preset historical health period is sampled to construct a voltage-current sample dataset for each terminal. ,in For the first Current values ​​at each sampling point For the first Voltage values ​​at each sampling point This represents the total number of sampling points; Based on the aforementioned sample dataset, a baseline model of voltage-current correlation between the terminals is established using a linear regression method. This baseline model is expressed as follows: ,in For the model slope parameter, For model intercept parameters, This refers to the load current of the wiring terminals; The residual sequence of each sampling point in the sample dataset relative to the baseline model is calculated using the following formula: ,in For the first Measured voltage values ​​at each sampling point; Calculate the standard deviation of the residual series ; The Mahalanobis distance of each sampling point is calculated using the following formula: ,in For the first The absolute value of the residuals at each sampling point; The goodness-of-fit index of the model is determined by the following formula: ,in This represents the mean of the voltage sample data; When the model fit index When the health status falls below the first preset threshold, the preset historical health period is reselected for modeling.

3. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 2, characterized in that, Step S2.2 includes the following steps: Get the current sampling time Real-time current value With real-time voltage value And based on the voltage-current correlation baseline model of the terminals. Calculate the predicted voltage value at the current moment. , The current sampling time The real-time current value; Calculate the residual at the current time step ; Get continuous residual sequence at each sampling time point ,in The preset sliding time window length; Calculate the mean of the residual sequence. Calculate the standard deviation of the residual sequence. ; Calculate the voltage sag slope deviation index using the formula. ,in The absolute value of the mean of the residual sequence. The standard deviation of the residual sequence is denoted as . The standard deviation of the residual sequence obtained when constructing the baseline model is given. The preset volatility weighting coefficient; When the voltage drop slope deviates from the exponential When the second preset threshold is exceeded, it is determined that the current real-time data deviates significantly from the baseline model.

4. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 2, characterized in that, Step S2.3 includes the following steps: Identify and obtain the target terminal block, which is the sub-healthy terminal block that is in a critical state of increased contact resistance but has not triggered a visible alarm; Based on the voltage drop slope deviation index of all terminals of the same model in the meter box, construct a set of deviation index datasets. ,in For the first The voltage drop slope deviation index of the same type of terminal block. This refers to the total number of terminals of the same type within the same meter box; The mean of the deviation index dataset within the same group is calculated using the following formula: ; Calculate the standard deviation of the same group of deviation index datasets, using the following formula: ; The lateral outlier coefficient of the target terminal block is calculated using the following formula: ,in The voltage drop slope deviation index of the target terminal; When the lateral outlier coefficient When the value exceeds the third preset threshold, the target terminal is determined to be a sub-healthy terminal that is in a critical state of increased contact resistance but has not triggered a visible alarm. Calculate the percentile ranking of the target terminal block in the same group of deviation index datasets. ; When the percentile ranking The lateral outlier coefficient is greater than the fourth preset threshold and When the value exceeds the third preset threshold, a first risk label is assigned to the target terminal block. when When the sample size is less than the fifth preset threshold, it is determined that the sample size of the terminal in the same group is insufficient. The longitudinal historical comparison is used instead of the horizontal comparison. The trend analysis is performed based on the historical deviation index sequence of the target terminal itself. When the voltage drop slope deviation index of multiple consecutive sampling periods shows a monotonically increasing trend and the current value exceeds the sixth preset threshold by a multiple of the historical average, the target terminal is given a first risk label.

5. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 1, characterized in that, Step S3 includes the following steps: The explicit electrical events include voltage sag / boost events, current loss events, and harmonic surge events; Record the occurrence time of the aforementioned explicit electrical events. and event types And extract the occurrence time of the explicit electrical event. Electrical characteristic data within the preset time window before and after; The event impact phase is determined based on the electrical characteristic data. ; When one of the following conditions is met: the phase voltage sag exceeds the seventh preset threshold, the phase current suddenly drops to near zero, or the phase harmonic content increases before and after the event exceeds the eighth preset threshold, the phase is determined to be the phase affected by the event. Based on the phase affected by the event and the time of the event, electrical topology information pre-stored in the knowledge graph is obtained and combined with the knowledge graph. A set of candidates with direct electrical connection with the phase affected by the event is retrieved from the knowledge graph. Obtain the time of occurrence of each candidate terminal in the candidate terminal block set during the dominant electrical event. The electrical characteristic data within the preset time window before and after the test are used to calculate the characteristic change of each candidate terminal. The calculation formula is: ,in The first time window before the event occurs The average electrical characteristic parameters of each candidate terminal; Within a preset time window after the event occurs The average electrical characteristic parameters of each candidate terminal; According to the formula Calculate the correlation index. Candidate terminals exceeding the ninth preset threshold are identified as suspected associated terminals; Obtain the geographical location and electrical path distance of the suspected associated terminals. According to the formula Calculate the overall association confidence score, where For preset weighting coefficients, The maximum path distance. For indicator variables; The terminal with the highest overall association confidence is identified as the terminal associated with the explicit electrical event, and it is associated with the event information and stored in the knowledge graph.

6. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 1, characterized in that, Step S4 includes the following steps: S4.1 Establish spatially related edges; S4.2 Calculate the time correlation coefficient; Step S4.1 includes the following steps: Obtain traffic vibration source information from the external environment data, and incorporate each traffic vibration source as an independent entity into the knowledge graph; Obtain the geographic coordinates of each terminal block bearing the first risk label. ; Calculate the first according to the formula. The traffic vibration source and the first Spacing between terminals bearing the first risk label ,in For the Earth's radius, and The first The latitude and longitude of each terminal block. and The first The latitude and longitude of each traffic vibration source; When the spatial distance When it is less than the tenth preset threshold, determine the first The terminal block is located at the first Within the influence range of a traffic vibration source; Calculate the first according to the formula. Comprehensive vibration impact index of each terminal block ,in The preset weighting coefficients and , The preset maximum influence distance, For the first Normalized value of the proportion of heavy vehicles among traffic vibration sources For the first Normalized values ​​of vibration intensity levels of individual traffic vibration sources; When the spatial distance When the value is less than the tenth preset threshold, it is the first in the knowledge graph. The traffic vibration source entity and the first Establish spatial association edges between the terminal block entities; Set attribute fields for the spatially associated edges; store the spatially associated edges and their attribute information in the knowledge graph, and update the association information field of the terminal block entity, adding the list of traffic vibration sources associated with it and the corresponding comprehensive vibration impact index.

7. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 6, characterized in that, Step S4.2 includes the following steps: Obtain traffic vibration source entities that have spatially associated edges with the target terminal block, and extract the set of peak traffic periods for heavy vehicles associated with the traffic vibration sources. Each peak period is defined by its start time. and end time definition; Obtain the occurrence time of the explicit electrical event. And convert it to the same time scale as the peak period; Calculate the first The time overlap between the peak period and the time of the event. The time correlation coefficient is calculated using the following formula: The calculated time correlation coefficient As attribute information, it is stored in the spatial association edge between the target terminal block and the traffic vibration source in the knowledge graph.

8. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 1, characterized in that, The determination conditions for the compound reasoning rule in step S5 include: The first condition is that the target terminal corresponding to the current loss event has the first risk label; The second condition is that the target terminal block and the traffic vibration source have a spatially related edge. The third condition is the temporal correlation coefficient between the occurrence time of the flow loss event and the peak period of heavy vehicle traffic at the traffic vibration source. Greater than the eleventh preset threshold; The fourth condition is that, after analyzing the waveform characteristics of the current loss event, it is confirmed that it is accompanied by a sudden increase in harmonic content or voltage flicker. When the first, second, third, and fourth conditions are met simultaneously, the risk attribution category of the current loss event is determined to be an environmental coupling failure; the determination result of the composite reasoning rule is used as the basis for updating the status label of the target terminal.

9. The method for early warning of inspection risks of traffic vibration coupled type power metering anomalies according to claim 1, characterized in that, Step S6 includes the following steps: When the risk attribution category of the power loss event is determined to be an environmental coupling failure based on the composite reasoning rule, an equipment maintenance work order is generated. An inspection work order is generated when the risk attribution category of the current loss event does not meet all the judgment conditions of environmental coupling failure and there are no other clear attributions, or when the characteristics of electricity theft are detected. Obtain the on-site verification results fed back after the completion of the processing work order; update the status label of the target terminal in the knowledge graph based on the on-site verification results; Based on the comparison between the on-site verification results and the reasoning conclusions, update the confidence parameter of the composite reasoning rule; The updated confidence parameters are stored as attributes of the composite inference rule in the knowledge graph.

10. A traffic vibration-coupled power metering anomaly inspection risk early warning system, characterized in that, include: The data acquisition module obtains power operation data and external environmental data at the target metering point; The terminal status diagnosis module constructs a voltage-current correlation baseline model for each terminal based on the preset historical health period data in the power operation data. The real-time power operation data is compared with the baseline model to calculate the voltage drop slope deviation index; based on the horizontal comparison of terminals of the same model in the same meter box, sub-healthy terminals that are in a critical state of increased contact resistance but have not triggered explicit alarms are identified, and the sub-healthy terminals are assigned a first risk label. An electrical event monitoring and location module monitors and captures explicit electrical events in the power operation data; When the explicit electrical event is captured, the event occurrence time and event type are recorded, and the associated target terminal is located; The environmental correlation analysis module materializes traffic vibration sources in external environmental data and incorporates them into a knowledge graph. It calculates the spatial distance between each traffic vibration source and the location of the terminal block with the first risk label, and establishes spatial correlation edges. It also performs a time-series comparison between the occurrence time of the explicit electrical event and the peak traffic hours of the corresponding traffic vibration source for heavy vehicles, and calculates the time correlation coefficient. The composite reasoning decision module constructs composite reasoning rules based on the spatial correlation edge, the temporal correlation coefficient, and the first risk label; and performs risk attribution determination on the current loss event in the explicit electrical event based on the composite reasoning rules to obtain the risk attribution category. The work order processing and feedback learning module generates differentiated work orders based on the risk attribution categories. The execution result of the processing work order is fed back to the knowledge graph to update the status label of the corresponding terminal block and the confidence parameter of the composite inference rule.

Images