High-low pressure linkage line loss comprehensive intelligent diagnosis system and method

Abstract

The application discloses a high-low voltage linkage line loss comprehensive intelligent diagnosis system and method, the system comprises a distributed storage and high-performance computing module, a data fusion and processing module, a multi-dimensional feature construction module, a high-low voltage linkage analysis module and a line loss intelligent diagnosis module; the method is based on the above system to carry out line loss comprehensive intelligent diagnosis, including real-time processing of multi-source data through the distributed storage and high-performance computing module; fusing bus-line-user file data and calculating line loss indicators; extracting time, space and electrical three-dimensional feature indicators; performing high-low voltage linkage analysis based on Pearson correlation coefficient and DTW distance; and adopting a system state model and an isolated forest anomaly detection model to fuse and output abnormal diagnosis results. The application is suitable for precise analysis and management of line loss in an intelligent power grid environment, and can improve the efficiency and quality of line loss management.

Description

Technical Field

[0001] This invention belongs to the field of power system line loss management technology, specifically, it relates to a comprehensive intelligent diagnostic system and method for line loss that links high and low voltage. Background Technology

[0002] In the daily operation of power systems, line loss has always been a key factor affecting the economic benefits and power quality of power supply companies. Traditional line loss management methods rely heavily on manual experience and simple statistical analysis, which have many limitations:

[0003] The analysis of high-voltage and low-voltage lines is often independent and cannot capture the correlation between the two.

[0004] Feature extraction has only one dimension and cannot fully reflect the complex influencing factors of line loss;

[0005] The diagnostic methods rely on a single indicator or simple threshold judgment, without fully integrating machine learning and deep learning algorithms. They are not adaptable enough to complex power consumption patterns and dynamically changing multi-level power grid environments, and cannot conduct in-depth analysis of multi-dimensional features. The accuracy and generalization ability of the models need to be improved, resulting in the inability to locate abnormal line losses in a timely and accurate manner.

[0006] With the advancement of smart grid construction, massive amounts of power data are constantly being generated, and traditional methods can no longer meet the needs for accurate diagnosis and effective management of line losses. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a comprehensive intelligent diagnostic system and method for line loss that links high and low voltage, applicable to the accurate analysis and management of line loss in smart grid environments, and capable of improving the efficiency and quality of line loss management.

[0008] To achieve the above objectives, the technical solution provided by the present invention is as follows:

[0009] According to a first aspect of the present invention, a comprehensive intelligent diagnostic system for line loss with high and low voltage linkage is proposed, comprising:

[0010] The distributed storage and high-performance computing module is used to realize distributed storage and real-time parallel processing of multi-source data;

[0011] The data fusion and processing module is used to build a dynamic archive connection system and calculate the line loss rate and bus unbalance rate.

[0012] The multi-dimensional feature construction module is used to construct a set of feature indicators in time, space and electrical dimensions from multiple levels such as voltage, current, power, power factor, electrical quantity and line loss.

[0013] The high- and low-voltage linkage analysis module is used to realize multi-level anomaly location of gateways, lines, dedicated and public transformer users and low-voltage users based on the Pearson correlation coefficient method and the dynamic time warping (DTW) algorithm.

[0014] The line loss intelligent diagnosis module is used to integrate the output results of the system state model and the isolated forest anomaly detection model to output anomaly diagnosis results.

[0015] Furthermore, the distributed storage and high-performance computing module includes:

[0016] Distributed storage units are used to store multi-source data based on HBase or Hive, including bus profiles, line profiles, high-voltage user profiles, transformer area profiles, junction lists, line junctions, daily current curve data for dedicated transformer users and low-voltage users; daily voltage curve data for line junctions, dedicated transformer users and low-voltage users; daily power curve data for line junctions, dedicated transformer users and low-voltage users; power factor data for line junctions, dedicated transformer users and low-voltage users; electricity consumption data for line junctions, dedicated transformer users and low-voltage users; and line loss data for line junctions, dedicated transformer users and low-voltage users.

[0017] High-performance computing units are used for real-time parallel processing of multi-source data based on Spark or Flink frameworks.

[0018] Furthermore, the data fusion and processing module includes:

[0019] The data fusion unit is used to build a dynamic file connection system for "busbar-line-dedicated and public transformer users-low voltage users";

[0020] The line loss index construction unit is used to construct the line loss calculation model and the bus level calculation model, and to calculate the input and output power and line loss rate of each distribution line.

[0021] The data preprocessing unit is used for outlier handling, missing value imputation, and data normalization.

[0022] The line loss indicator construction unit includes:

[0023] The line loss calculation subunit is used to calculate line loss according to the following formula:

[0024] Power supply = forward power at line junctions + reverse power at public transformer junctions + reverse power at dedicated transformer users + reverse power for high-voltage office use + power connected to the grid by high-voltage distributed energy sources;

[0025] Electricity consumption = Forward electricity consumption of dedicated transformer users + Forward electricity consumption of public transformer substations + Forward electricity consumption of high-voltage office equipment + Reverse electricity consumption of line substations;

[0026] Line loss rate = (Power supply - Power consumption) / Power supply × 100%;

[0027] The busbar imbalance calculation sub-unit is used to calculate the busbar imbalance rate according to the following formula:

[0028] Busbar imbalance rate = (busbar power supply - busbar power consumption) / busbar power supply × 100%.

[0029] Furthermore, the multi-dimensional feature construction module includes:

[0030] The time dimension feature unit is used to extract load change features at different time scales, including the mean, standard deviation, maximum and minimum values ​​of electricity and line loss for different periods, and the daily load curve fluctuation coefficient.

[0031] Spatial dimension feature unit, used to calculate the spatial conductivity characteristics of high and low voltage networks, including power similarity coefficient, power loss similarity coefficient and line loss rate similarity coefficient;

[0032] Electrical dimension feature units are used to calculate voltage level, fluctuation, unbalance, and phase deviation; current level, fluctuation, unbalance, phase deviation, and trend correlation; power level, fluctuation, and total differential; power factor level and fluctuation index; and voltage, current, and power cross-features at the same time section.

[0033] Furthermore, the high-low voltage linkage analysis module is configured as follows: First, it detects bus line loss anomalies based on the bus line loss linkage analysis model and line loss anomaly identification business rules; second, it locates the abnormal medium-voltage lines based on the bus line loss linkage analysis model and line line loss linkage analysis model; then, it locks the abnormal dedicated and public transformer users according to the line line loss linkage analysis model; next, it outputs a list of suspected electricity theft clues for low-voltage users according to the public transformer area line loss linkage analysis model; finally, it realizes multi-level anomaly location of gateways, lines, dedicated and public transformer users, and low-voltage users based on the multi-level linkage analysis model formed by the bus line loss linkage analysis model, line line loss linkage analysis model, and public transformer area line loss linkage analysis model.

[0034] The bus line loss linkage analysis model is configured as follows:

[0035] In response to the detection of abnormal bus line loss, the analysis of the upper and lower levels of bus line loss is triggered;

[0036] If the main transformer loss is analyzed upwards, and there is a reverse coupling relationship between the main transformer loss and the bus line loss, then the low-voltage side of the main transformer is initially located as abnormal.

[0037] The analysis of line loss is linked downwards. If there is a reverse coupling relationship between the line loss and the bus line loss, the abnormality of the line junction can be initially located.

[0038] The route loss linkage analysis model is configured as follows:

[0039] In response to the detection of abnormal line loss or the receipt of abnormal medium-voltage line diagnostic results from the bus line loss linkage analysis model, line loss linkage analysis is triggered.

[0040] Downlink analysis of public transformer line losses and dedicated transformer power consumption;

[0041] If there is a reverse coupling relationship between the line loss and the transformer line loss, the transformer gate anomaly can be initially identified.

[0042] If there is a strong correlation between line loss and the electricity consumption of the dedicated transformer, the abnormality of the dedicated transformer meter can be initially identified.

[0043] The transformer substation line loss linkage analysis model is configured as follows:

[0044] In response to the detection of abnormal line loss in the transformer area or the receipt of abnormal public transformer user diagnosis results from the line loss linkage analysis model, the transformer area line loss linkage analysis is triggered.

[0045] Analyze the electricity consumption of low-voltage users by linking the analysis to the lower voltage level;

[0046] If there is a strong correlation between line loss in the transformer area and electricity consumption of low-voltage users, the initial location of the problem is an abnormality in the metering of low-voltage users.

[0047] Correlation analysis was performed using both Pearson correlation coefficient and dynamic time-warped (DTW) distance; the presence of strong negative or strong positive correlations was used to identify anomalies.

[0048] Furthermore, the reverse coupling relationship means that when an anomaly occurs at a certain metering point, the line loss of its superior level and the line loss of the level where the point is located exhibit a correlation characteristic of mutual inversion and reverse fluctuation, that is, the line loss on one side increases while the line loss on the other side decreases.

[0049] Furthermore, the intelligent line loss diagnosis module includes:

[0050] The system state model unit is used to construct a line system state model based on energy conservation and to calculate the metering point error rate.

[0051] The isolated forest anomaly diagnosis unit is used to identify anomalous data by constructing an isolated forest anomaly detection model;

[0052] The model fusion unit is used to integrate the outputs of the system state model and the isolated forest anomaly detection model to generate high-confidence anomaly clues and secondary clues.

[0053] The system state model unit is constructed based on the following formula:

[0054] ;

[0055] in For the first Antenna power supply; For the first Heavenly Electricity consumption at each metering point ,express One measurement point; For the first Error rate of each measurement point; For the first Daily wire loss; For the first The copper loss of the transformer is a variable loss. This refers to the iron loss of the transformer, i.e., the constant loss. For model residuals;

[0056] The isolated forest anomaly diagnosis unit achieves anomaly detection through the following steps:

[0057] Create an isolated forest by constructing multiple isolated trees;

[0058] Calculate the path length of the sample in the isolated tree;

[0059] Anomaly scores are calculated based on path length to identify anomalous samples.

[0060] The model fusion unit generates anomaly clues through the following steps:

[0061] Consensus anomaly samples from the system state model and the isolated forest anomaly detection model are extracted as high-confidence cues;

[0062] Non-consensus anomaly samples are subjected to archival standardization verification, new energy user characteristic filtering, and temporal continuity verification to generate secondary clues.

[0063] According to a second aspect of the present invention, a comprehensive intelligent diagnostic method for line loss based on the system is proposed, comprising the following steps:

[0064] Step S1: Process multi-source data in real time through distributed storage and high-performance computing modules;

[0065] Step S2: Integrate bus-line-user profile data and calculate line loss indicators;

[0066] Step S3: Extract time, space, and electrical three-dimensional feature indicators;

[0067] Step S4: Perform high-low pressure linkage analysis based on Pearson correlation coefficient and DTW distance;

[0068] Step S5: The system state model and the isolated forest anomaly detection model are fused together to output the anomaly diagnosis results.

[0069] The high-low voltage linkage analysis in step S4 specifically includes:

[0070] When the Pearson correlation coefficient r between the line loss and the transformer substation line loss is less than -0.7 and the DTW distance is less than the preset threshold, the transformer metering point is determined to be abnormal.

[0071] When the Pearson correlation coefficient |r| between line loss and electricity consumption of dedicated transformer users is greater than 0.8, the dedicated transformer user is considered to be abnormal.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0073] This invention significantly improves the accuracy, real-time performance, and intelligence of power system line loss management through innovative high- and low-voltage linkage analysis, multi-dimensional feature extraction, distributed real-time computing, and a comprehensive intelligent diagnostic model for line losses. Specific advantages are as follows:

[0074] 1. This invention integrates and analyzes data from high-voltage and low-voltage lines, and establishes a high-low voltage linkage analysis model using the Pearson correlation coefficient method and the Dynamic Time Warping (DTW) algorithm. This enables collaborative diagnosis of line loss problems between high-voltage and low-voltage lines, which differs from the traditional single voltage level analysis mode and improves the accuracy of anomaly location.

[0075] 2. This invention employs multi-dimensional feature indicators to enhance model adaptability;

[0076] 3. This invention utilizes distributed storage and high-performance computing technologies to construct a real-time analysis model, thereby improving the ability to identify line loss fluctuation characteristics and provide intelligent early warning.

[0077] 4. This invention optimizes accuracy and generalization ability through a comprehensive intelligent diagnostic model for line loss. Attached Figure Description

[0078] Figure 1 This is a flowchart of the integrated intelligent diagnosis method for line loss with high and low voltage linkage in an embodiment of the present invention. Detailed Implementation

[0079] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, this invention is not limited to the following embodiments, and specific implementation methods can be determined according to the technical solutions of this invention and actual circumstances. To avoid obscuring the essence of this invention, well-known methods, processes, and procedures are not described in detail.

[0080] like Figure 1 As shown, the integrated intelligent diagnostic method for line loss involving high and low voltage linkage includes distributed storage and high-performance computing, fusion and processing of multi-source data, construction of multi-dimensional features, analysis of high and low voltage linkage analysis model, and construction of integrated intelligent diagnostic model for line loss.

[0081] (I) Distributed Storage and High-Performance Computing

[0082] This project constructs a real-time big data analysis and processing model based on distributed storage (such as HBase and Hive) and high-performance computing (such as Spark and Flink) to achieve efficient storage and parallel processing of data from multiple systems. By deeply mining the value of data from multiple systems, it performs real-time processing on massive amounts of high-frequency data, providing a real-time and efficient data foundation for subsequent data fusion processing and the construction of multi-dimensional feature indicators. It should be noted that HBase is a distributed, column-oriented open-source database. Hive is a data warehouse analysis system built on Hadoop, providing rich SQL query methods for analyzing data stored in the Hadoop Distributed File System. Spark is an open-source distributed computing system developed by the Apache Software Foundation. Flink is an open-source stream processing framework developed by the Apache Software Foundation, with its core using the Java and Scala programming languages, supporting distributed data stream processing. HBase, Hive, Spark, and Flink are all existing technologies.

[0083] Specifically, the real-time big data analysis and processing model based on distributed storage and high-performance computing includes:

[0084] Distributed storage units are configured to use HBase or Hive to achieve efficient storage of multi-source system data;

[0085] High-performance computing unit, integrating Spark or Flink framework, to achieve parallel real-time processing of massive high-frequency data;

[0086] The distributed storage unit specifically includes:

[0087] The data sharding sub-unit horizontally shards the original data according to timestamps and business dimensions;

[0088] The compressed coding subunit uses a column family compression algorithm to reduce storage space usage;

[0089] The metadata management subunit maintains global consistency between the data table structure and partition information.

[0090] The high-performance computing unit achieves:

[0091] Based on Spark Streaming's micro-batch processing mode, a dynamic window size adjustment mechanism is configured;

[0092] Based on Flink's stream processing model, it supports event temporal semantics and state consistency guarantees.

[0093] (II) Fusion and processing of multi-source data

[0094] 1. Data Fusion

[0095] By leveraging the file linkage relationships between the integrated application system and the data acquisition system, a dynamic file connectivity system is constructed, linking "busbars - lines - dedicated and public transformer users - low-voltage users." This integrates files and operational data distributed across different systems, ensuring data continuity and integrity. Specific acquisition path:

[0096] Obtain busbar files, line files, high-voltage user files, transformer area files, and a list of junctions from the integrated system;

[0097] The system acquires the following data: customer profiles, operating electricity meter information, metering point relationships, and metering point profiles; daily current curve data for line junctions, dedicated transformer users, and low-voltage users; daily voltage curve data for line junctions, dedicated transformer users, and low-voltage users; daily power curve data for line junctions, dedicated transformer users, and low-voltage users; power factor data for line junctions, dedicated transformer users, and low-voltage users; electricity consumption data for line junctions, dedicated transformer users, and low-voltage users; and line loss data for line junctions, dedicated transformer users, and low-voltage users.

[0098] Based on archival data, a dynamic interconnected system is constructed, connecting "busbars - lines - dedicated and public transformer users - low-voltage users," linking data from the integrated system and the data acquisition system.

[0099] It should be noted that dedicated transformers and public transformers include dedicated transformers and public transformers. Dedicated transformers and public transformers are two power supply modes in the power system and belong to existing technologies.

[0100] 2. Construction of Line Loss Indicators After Integration

[0101] (1) Route loss calculation model: Based on the principle of energy conservation, a route loss calculation model is constructed, and the calculation formula is as follows:

[0102] 1) Power supply = forward power at line junction + reverse power at public transformer junction + reverse power for dedicated transformer users (non-distributed energy grid connection) + reverse power for high-voltage office use + power supply to high-voltage distributed energy grid connection;

[0103] 2) Electricity consumption = Forward electricity consumption of dedicated transformer users + Forward electricity consumption of public transformer substations + Forward electricity consumption of high-voltage office users + Reverse electricity consumption of line substations;

[0104] 3) Line loss rate = (Power supply - Power consumption) / Power supply × 100%;

[0105] By acquiring line-to-line transformation relationships, line junction electricity data, and junction forward / reverse relationships, the input and output electricity and line loss rate of each distribution line are calculated. The "line-to-line transformation relationship" is obtained from the data acquisition system and the integrated system, detailing the file relationships between lines and dedicated / public transformers. The "junction forward / reverse relationship" is determined by distinguishing the forward / reverse relationships of junction meters using fields in the corresponding tables of the data acquisition system. Additionally, the data acquisition system has corresponding fields in its tables to distinguish the following electricity data: forward electricity at line junctions, reverse electricity at public transformer junctions, reverse electricity for dedicated transformer users, reverse electricity for high-voltage office use, and electricity fed into the grid by high-voltage distributed energy sources; forward electricity for dedicated transformer users, forward electricity at public transformer junctions, forward electricity for high-voltage office use, and reverse electricity at line junctions.

[0106] (2) Busbar imbalance calculation model: Establish a busbar imbalance calculation model, based on the fact that the power flowing into the busbar is the power supply and the power flowing out of the busbar is the power consumption. Statistically summarize the input and output power information of the busbar. The calculation formula is as follows:

[0107] Busbar imbalance rate = (Busbar power supply - Busbar power consumption) / Busbar power supply × 100%;

[0108] By acquiring the busbar files and the list of busbar connected equipment, the system determines the key operating data of each component on the busbar, calculates the busbar imbalance, and thus monitors the busbar's operating status.

[0109] 3. Data Preprocessing

[0110] (1) Outlier handling: The statistical IQR method is used to identify abnormal data caused by unstable acquisition equipment, system failure or human error in recording. These abnormal data are characterized by sudden increases or decreases and significant irregular changes compared with previous and subsequent data. Outliers are removed or corrected. At the same time, data smoothing techniques (such as moving average method and exponential smoothing method) are used to reduce noise interference and improve data reliability.

[0111] (2) Missing value handling: For missing data caused by line faults, equipment maintenance or data transmission problems, the K nearest neighbor interpolation method is selected to fill the missing data according to the data characteristics, or data with similar periods in historical data is used to replace the missing data to ensure the integrity of the data.

[0112] (3) Data Conversion and Normalization: Through keyword recognition and primary key recognition, the archives, topology, and operational data of multiple source systems such as scheduling, collection, and marketing are integrated and merged to unify the data format; the load data is normalized to eliminate the influence of dimensions and orders of magnitude. The calculation formula is as follows:

[0113] ;

[0114] in, The original data, The minimum value of the data. For the maximum value of the data, Normalized data makes different features comparable, facilitating subsequent data analysis and processing.

[0115] (III) Construction of Multi-Dimensional Features

[0116] A feature index set is constructed from multiple levels, including voltage, current, power, power factor, electrical quantity, and line loss.

[0117] 1. Time Dimension Features: Extract load variation characteristics at different time scales, including the mean, standard deviation, maximum, and minimum values ​​of electricity consumption and line loss for different periods, as well as the daily load curve fluctuation coefficient. Daily Load Curve Fluctuation Coefficient The calculation formula is:

[0118] ;

[0119] in For a moment The load power; The number of sampling points per day; This refers to the point with the highest load power among the sampling points throughout the day. This refers to the point with the lowest load power among the sampling points in a day.

[0120] 2. Spatial Dimension Characteristics: The spatial conductivity characteristics of high and low voltage networks are calculated, including the similarity coefficients of electrical loads, electrical losses, and line loss rates at different voltage levels. The similarity between conductive load data is calculated using the Pearson correlation coefficient. Specifically, the similarity coefficients of electrical loads, electrical losses, and line loss rates at different voltage levels are calculated using the Pearson correlation coefficient. For detailed calculation procedures of the similarity coefficients of electrical loads, electrical losses, and line loss rates at different voltage levels, please refer to the "Pearson Correlation Coefficient Calculation Formula".

[0121] 3. Electrical Dimension Characteristics: These include voltage level, fluctuation, unbalance, and phase deviation; current level, fluctuation, unbalance, phase deviation, and trend correlation; power level, fluctuation, and total differential; power factor level and fluctuation indicators; and the cross-characteristics of voltage, current, and power at the same time cross-section. The unbalance calculation formula is: Unbalance = (Maximum value - Minimum value) / Maximum value. All of the above characteristics are existing technologies and will be briefly explained below without further detail.

[0122] (1) Voltage-related characteristics

[0123] Level (amplitude): The deviation between the nominal voltage (e.g., 220V, 380V) and the actual measured value, reflecting the stability of the power supply.

[0124] Fluctuation: Voltage changes over time (such as transient fluctuations or periodic fluctuations), commonly measured by standard deviation or peak-to-peak value.

[0125] Unbalance: The degree of asymmetry in the amplitude or phase of three-phase voltage (such as negative sequence voltage ratio), which affects motor efficiency.

[0126] Phase deviation: The deviation of the voltage phase angle from the ideal value (such as a 120° deviation between three phases).

[0127] (2) Current-related characteristics

[0128] Level (amplitude): The magnitude of the load current, which is directly related to the equipment load rate.

[0129] Fluctuation: Current change rate (such as starting current, harmonic current), reflecting the dynamic characteristics of the load.

[0130] Unbalance: Asymmetrical three-phase current (such as due to excessive single-phase load) may cause overheating.

[0131] Phase deviation: the phase difference between current and voltage (power factor angle) or the phase asymmetry of three-phase current.

[0132] Trend correlation: Synchronization of current changes in multiple loops (such as the consistency of current in parallel equipment), used for fault location.

[0133] (3) Power-related characteristics

[0134] Level: Absolute values ​​of active power (P), reactive power (Q), and apparent power (S).

[0135] Fluctuation: the rate of change in power (such as fluctuations in the output of new energy sources).

[0136] Total differential: The difference between input and output power (such as the unbalanced power in transformer differential protection).

[0137] (4) Power factor characteristics

[0138] Level: The ratio of active power to apparent power, reflecting energy utilization efficiency.

[0139] Fluctuation: Dynamic changes in power factor (such as changes caused by motor start-up and shutdown).

[0140] (5) Cross-section characteristics under time section

[0141] Voltage-current phase relationship: impedance characteristic analysis (such as capacitive / inductive load determination).

[0142] Voltage-power correlation: The response of active / reactive power to voltage dips (e.g., low voltage ride-through capability).

[0143] Current-power trend: Synchronous changes in current and power (such as the correlation between overload and power exceeding limits).

[0144] (iv) High and low voltage linkage analysis

[0145] Based on the archives and data of the electricity consumption information collection system and the integrated electricity consumption and line loss platform, the system first identifies busbar line loss anomalies using the busbar line loss linkage analysis model and line loss anomaly identification business rules. Secondly, it locates the abnormal medium-voltage lines using the busbar line loss linkage model and the line loss linkage analysis model. Medium-voltage lines refer to the power lines connecting the substation busbar and the distribution transformer, which are the main transmission channels in the distribution network. Then, it identifies the abnormal distribution transformer users based on the line loss linkage analysis model. Next, it outputs a list of suspected electricity theft clues for low-voltage users based on the distribution transformer area line loss linkage analysis model. Finally, based on the multi-level linkage analysis model formed by the busbar line loss linkage analysis model, the line loss linkage analysis model, and the distribution transformer area line loss linkage analysis model, it achieves multi-level anomaly location at 10kV points, lines, distribution transformers, and low-voltage users, providing a basis for multi-level linkage anomaly management.

[0146] Line loss anomaly identification business rules: For bus line loss, medium voltage line loss and transformer area line loss, based on expert experience (such as transformer area line loss higher than 10% or lower than -1%) and theoretical line loss data (such as the absolute value of the deviation between theoretical line loss and line loss rate greater than 2), line loss anomalies are judged, and high negative loss, abnormal fluctuation and other line loss anomalies are initially identified and located.

[0147] Busbar loss linkage analysis model: Based on the preliminary identification of busbar loss anomalies, trigger the linkage analysis between the upper and lower levels of busbar loss; the upward linkage analysis analyzes the main transformer loss, and if there is a reverse coupling relationship between the main transformer loss and the busbar loss, the anomaly of the low-voltage side of the main transformer is initially located; the downward linkage analysis analyzes the line loss, and if there is a reverse coupling relationship between the line loss and the busbar loss, the anomaly of the line is initially located.

[0148] Line loss linkage analysis model: Based on the preliminary identification of abnormal line losses and the abnormal medium-voltage lines diagnosed by the downward linkage analysis of bus line losses, the line loss linkage analysis is triggered; the downward linkage analysis analyzes the public transformer line loss and the dedicated transformer power consumption. If there is a reverse coupling relationship between the line loss and the public transformer line loss, the abnormality of the public transformer gate is initially located. If there is a strong correlation between the line loss and the dedicated transformer power consumption, the abnormality of the dedicated transformer meter is initially located.

[0149] The reverse coupling phenomenon refers to the situation where, when an anomaly occurs at a metering point, the line loss at the upstream level and at that metering point level exhibits a change where "one side increases while the other decreases." For example, if the public transformer metering point under-meters, the power loss in the distribution area decreases. However, if the power supply remains constant, a decrease in electricity sales leads to an increase in line power loss.

[0150] Transformer Area Line Loss Linkage Analysis Model: Based on the preliminary identification of abnormal transformer area line losses and the abnormal public transformer users diagnosed by the downward linkage analysis of line losses, the transformer area line loss linkage analysis is triggered; the downward linkage analysis of low-voltage user electricity consumption is performed, and if there is a strong correlation between transformer area line loss and low-voltage user electricity consumption, the abnormality of low-voltage user metering is initially located.

[0151] Based on the energy transmission principle that the output power of the upper-level gate is the input power of the lower-level gate, if there is an anomaly at the public transformer metering point, such as under-metering, the line loss of the public transformer area will decrease, and when correlated upwards to the 10kV line loss, the line loss will increase. Given this theory, based on 10kV line loss data and public transformer area line loss data, a linkage analysis is performed on line loss and public transformer area line loss to calculate their correlation. If a strong negative correlation exists (in statistics, the correlation coefficient is generally defined as -0.7), an anomaly at the public transformer metering point is preliminarily identified. Furthermore, if a public-private transformer user is abnormal, their electricity consumption will show a strong correlation with line loss (in statistics, this is generally defined as 0.7; to screen for strong correlations, the threshold is increased to 0.8). Based on this, using line loss and public-private transformer user electricity consumption data, an abnormal public-private transformer user is preliminarily analyzed and diagnosed. The Pearson correlation coefficient method and the Dynamic Time Warping (DTW) algorithm are used to calculate the correlation between line loss and public-private transformer user electricity consumption, and the correlation between line loss and public transformer area line loss.

[0152] 1. Pearson correlation coefficient The calculation formula is:

[0153] ;

[0154] in, Corresponding electricity consumption data for dedicated and public transformer users; Corresponding route line loss data, For the first sky, For a total Data from the day, For user battery level Daily average For line loss in The average value over the days.

[0155] 2. DTW (Dynamic Time Warping) Distance Calculation:

[0156] DTW (Dynamic Time Warping) distance calculation addresses the challenge in power line loss analysis scenarios where traditional Euclidean distance struggles to accurately measure the similarity between different time series (such as load curves and line loss data series) due to potential time axis misalignment or scale differences. The DTW algorithm, through dynamic programming, allows for non-linear alignment of time series along the time axis, effectively resolving the issue of traditional Euclidean distance's inaccurate measurement of similarity. Given two time series... and ,in This represents the time series of electricity consumption by dedicated and public transformer users. This represents the time series of line losses. For dedicated public transformer users in the first Daily electricity consumption data Indicates the line at the 1st Daily line loss data. This is achieved by constructing a cumulative distance matrix. Its elements The calculation formula is as follows:

[0157] ;

[0158] In the formula: For local distance measurement, , The difference is amplified using the squared error method; the initial condition is... Curved paths in the distance matrix The range reached is called the curved window. The final distance metric is defined as:

[0159] ;

[0160] in, For curved paths The length of this value is determined by backtracking the class distance matrix. The smaller the calculated DTW distance value, the higher the similarity between the two time series.

[0161] Based on the above calculations, the correlation analysis module (comprising Pearson correlation coefficient calculation and DTW distance calculation) is used to perform Pearson correlation coefficient and DTW distance calculations on the load and line loss rate time series of each public and private transformer user and distribution area with the line loss series. This initially identifies abnormal public and private transformer users with a strong correlation to line loss, providing data support for subsequent targeted investigations into electricity theft, metering equipment failures, and abnormal line loss causes. When the Pearson correlation coefficient r between line loss and public transformer distribution area line loss is less than -0.7 and the DTW distance is less than a preset threshold, the public transformer metering point is determined to be abnormal.

[0162] When the Pearson correlation coefficient |r| between line loss and electricity consumption of dedicated transformer users is greater than 0.8, the dedicated transformer user is considered to be abnormal.

[0163] (V) Intelligent Algorithm for Line Loss Diagnosis

[0164] Based on the initial dedicated transformer clues output from the high and low voltage linkage analysis, a comprehensive intelligent diagnostic model for line loss is constructed based on the electrical principle of energy conservation and machine learning algorithms (the comprehensive intelligent diagnostic model for line loss is a model that integrates the system state model and the isolated forest anomaly diagnostic model) to perform in-depth analysis and diagnosis of the system data to which the clue data belongs.

[0165] 1. System State Model

[0166] Taking a 10kV line as the system and dedicated and public transformer metering points as the analysis objects, based on the principle of conservation of power supply, sales and losses of the line, using the power supply of public and dedicated transformers and line loss data, and using machine learning technology, a system state model is constructed, the model parameters are calculated, and the state parameters of each metering point, i.e., the error value, are obtained.

[0167] Constructing a system state model based on energy conservation:

[0168] ;

[0169] in: For the first Antenna power supply; For the first Heavenly Electricity consumption at each metering point ,express One measurement point; For the first Error rate of each measurement point; For the first Daily wire loss; For the first The copper loss of the transformer is the heat loss generated by the winding resistance during transformer operation. It is proportional to the square of the load current and belongs to variable loss. The iron loss of the distribution transformer (no-load loss) is the fixed loss generated by the iron core during the operation of the transformer, that is, the constant loss. This represents the model residuals.

[0170] Based on the maximum allowable error parameter of the meter (e.g., ±0.2% for a 0.2S class meter), a threshold discrimination matrix for the metering point status parameter is constructed. The error rate of each metering point is calculated through the system status model and dynamically compared with the error limit of the corresponding meter accuracy class. When the metering point status parameter value exceeds the upper limit of the allowable error (e.g., positive error > +0.2% or negative error < -0.2%), an anomaly marking mechanism is automatically triggered, and a list of meters exceeding the threshold is output as a diagnostic clue for the system status model.

[0171] 2. Isolated Forest Anomaly Diagnostic Model

[0172] In anomaly detection tasks, commonly used algorithms include Isolation Forest, clustering methods, and filtering mechanisms based on practical business rules. Among these, Isolation Forest is widely used in industry due to its efficiency and strong ability to identify global anomalies; clustering methods rely on distance metrics and data distribution characteristics; while methods based on business rules are often limited by the accuracy of threshold settings. Given the characteristics of the analyzed data—multi-dimensional data and a low proportion of anomaly samples—the Isolation Forest algorithm is chosen as the primary anomaly detection tool.

[0173] Algorithm Principle: Isolation Forest is an unsupervised anomaly detection algorithm based on ensemble learning. Its core idea is to "isolate" data points by constructing a series of random binary trees. Because anomalous data are fewer in number and significantly different from normal data in the feature space, they are easier to isolate quickly. Specifically, anomalous points are usually closer to the root node in the tree structure; normal points require more partitioning steps to be isolated, i.e., they are located in deeper leaf nodes. By constructing multiple isolated trees to form a "forest" and combining the results of each tree, the robustness and stability of the model can be effectively improved.

[0174] The core mechanism and mathematical expression of the algorithm are as follows: Let the dataset be... Each sample , indicating a 3D eigenvectors.

[0175] (1) The process of constructing an isolated tree

[0176] The construction process for each isolated tree is as follows:

[0177] 1) Randomly select a feature dimension;

[0178] 2) Randomly select a split value between the maximum and minimum values ​​of this dimension;

[0179] 3) Divide the data into left and right subsets;

[0180] 4) Recursively repeat the above steps until the current node has only one sample or the maximum depth limit is reached.

[0181] (2) Isolated Forest Integration

[0182] Build Each tree is an isolated tree, forming a forest. The training samples for each tree are obtained through random sampling without replacement to ensure the robustness of the model.

[0183] (3) Path length and anomaly score

[0184] 1) Path length. For a sample, its path length in the isolated tree is denoted as... This refers to the number of edges traversed from the root node to the leaf node containing the sample, reflecting the ease with which it is "isolated." Outlier samples, due to their significant feature differences, typically have shorter path lengths. (Definition of sample) The average path length is: ,in This indicates the number of isolated trees in an isolated forest.

[0185] 2) Anomaly Score. The formula for calculating the anomaly score is: ,in It is the normalization factor, representing the normalization factor for a given sample size. The expected value of the average path length at time is calculated using the following formula: .

[0186] 3) Abnormal judgment criteria

[0187] when When the path length is shorter, it indicates that the sample is highly likely to be an outlier; when... When this time, it indicates that the sample is a normal point; generally, it is... This serves as a threshold for determining anomalies.

[0188] The isolated forest anomaly detection model is applied to the constructed multi-dimensional feature index data to output a list of meters with high anomaly scores and a large number of abnormal days for subsequent analysis and early warning.

[0189] 3. Model fusion output

[0190] To improve the accuracy and coverage of intelligent line loss diagnosis, a model fusion strategy is adopted to combine the output results of the system state model and the isolated forest anomaly detection model:

[0191] (1) High-confidence clues

[0192] Set intersection operation is used to extract bi-model consensus anomaly samples. That is, when the error rate of the measurement points calculated by the system state model exceeds the allowable threshold and the anomaly score output by the isolated forest anomaly detection model is ≥0.8, it is judged as a high-confidence anomaly clue.

[0193] (2) Secondary clues

[0194] For the non-consensus anomaly set output by the dual models, a three-layer data cleaning mechanism is constructed:

[0195] 1) Verification of archival standardization: By checking archival data and collected curve data, we can eliminate users with mismatched three-phase three-wire / four-wire systems (such as 3×100V being mistakenly archived as 3×220V) and abnormal users with confused voltage levels, so as to ensure the accuracy of the basic data of the analysis object;

[0196] 2) New energy user feature filtering: Identify photovoltaic grid-connected users based on current direction features, and independently mark negative current samples to avoid misjudgment caused by bidirectional energy flow;

[0197] 3) Temporal continuity verification: A sliding window is used to perform temporal consistency verification on the remaining abnormal samples to eliminate false anomalies caused by occasional data fluctuations.

[0198] The remaining samples after the above filtering constitute a secondary clue set, providing data support for manual review.

[0199] This fusion strategy combines the advantages of system state assessment and automatic model identification, improving the accuracy and practicality of overall anomaly identification.

[0200] The high- and low-voltage linkage analysis model in this invention proposes to integrate and analyze data from high-voltage and low-voltage lines, and establish a high- and low-voltage linkage analysis model using the Pearson correlation coefficient method and the DTW algorithm to achieve collaborative diagnosis of line loss problems between high- and low-voltage lines, which is different from the traditional single voltage level analysis mode.

[0201] The multi-dimensional feature index set in this invention constructs a multi-dimensional feature extraction framework that includes time, space, and electrical dimensions, providing rich data input for anomaly diagnosis and improving the model's adaptability to complex power consumption scenarios.

[0202] The distributed real-time analysis model in this invention utilizes distributed storage and high-performance computing technologies to achieve real-time processing and intelligent early warning of massive amounts of data, significantly improving data processing efficiency and the real-time performance of line loss monitoring, thus meeting the needs of lean governance.

[0203] The line loss integrated intelligent diagnostic model in this invention integrates the electrical principle system state model and the isolated forest machine learning algorithm to construct an intelligent diagnostic model. By leveraging the advantages of different algorithms, it achieves comprehensive diagnosis and accurate identification of anomalies at various voltage levels, thereby improving the model's accuracy and generalization ability.

[0204] Compared with the prior art, the advantages of the present invention are as follows:

[0205] 1. More efficient data fusion and processing: Through multi-source data integration and cleaning technology, data quality is improved, providing a more reliable data foundation for line loss analysis. Compared with existing technologies, it has more data dimensions and higher processing accuracy.

[0206] 2. More accurate anomaly location: By utilizing high and low voltage linkage analysis and multi-dimensional feature extraction, combined with advanced algorithms, abnormal lines and users can be located more accurately, solving the problem of low location efficiency of traditional methods.

[0207] 3. Enhanced real-time analysis capabilities: The real-time analysis model, built upon distributed storage and high-performance computing, enables real-time monitoring and intelligent early warning of line loss fluctuations, meeting the needs of modern power grid real-time management.

[0208] 4. Superior Model Performance: The integrated intelligent diagnostic model for line loss incorporates multiple algorithms, improving the model's accuracy and generalization ability. It is more adaptable to complex power environments and provides better diagnostic results compared to existing models.

Claims

1. A high-low pressure linkage line loss comprehensive intelligent diagnosis system, characterized in that, Comprise: A distributed storage and high-performance computing module for realizing distributed storage and real-time parallel processing of multi-source data; A data fusion and processing module for constructing a dynamic file through system and calculating line loss rate and bus imbalance rate; A multi-dimensional feature construction module for constructing time dimension, space dimension and electrical dimension feature index set from voltage, current, power, power factor, electricity and line loss in multiple aspects; A high-low voltage linkage analysis module for realizing multi-level abnormal positioning of gateway, line, special public variable user and low voltage user based on Pearson correlation coefficient method and dynamic time warping (DTW) algorithm; An intelligent line loss diagnosis module for outputting abnormal diagnosis results by comprehensively combining the output results of system state model and isolated forest anomaly detection model. The high-low voltage linkage analysis module is configured to: first, discover bus line loss anomaly based on bus line loss linkage analysis model and line loss anomaly identification business rules; second, locate abnormal medium voltage line based on bus line loss linkage analysis model and line loss linkage analysis model; third, lock abnormal special public variable user according to line loss linkage analysis model; fourth, output low voltage user suspected electricity stealing clue list according to public variable area line loss linkage analysis model; and fifth, realize multi-level abnormal positioning of gateway, line, special public variable user and low voltage user based on multi-level linkage analysis model formed by bus line loss linkage analysis model, line loss linkage analysis model and public variable area line loss linkage analysis model. The bus line loss linkage analysis model is configured to: In response to identifying bus line loss anomaly, trigger upper and lower linkage analysis of bus line loss; Analyze main transformer loss upwards, if there is a reverse coupling relationship between main transformer loss and bus line loss, preliminarily locate main transformer low voltage side gateway anomaly; Analyze line loss downwards, if there is a reverse coupling relationship between line loss and bus line loss, preliminarily locate line gateway anomaly. The line loss linkage analysis model is configured to: In response to identifying line loss anomaly or receiving abnormal medium voltage line diagnosis results from the bus line loss linkage analysis model, trigger line loss linkage analysis; Analyze public variable line loss and special variable electricity downwards; If there is a reverse coupling relationship between line loss and public variable line loss, preliminarily locate public variable gateway anomaly; If there is strong correlation between line loss and special variable electricity, preliminarily locate special variable meter abnormality. The public variable area line loss linkage analysis model is configured to: In response to identifying area line loss anomaly or receiving abnormal public variable user diagnosis results from the line loss linkage analysis model, trigger area line loss linkage analysis; Analyze low voltage user electricity downwards; If there is strong correlation between area line loss and low voltage user electricity, preliminarily locate low voltage user meter abnormality. Correlation analysis is performed using Pearson correlation coefficient and dynamic time warping (DTW) distance; wherein, existence of strong negative correlation or strong positive correlation is used to determine anomaly.

2. The high-low voltage linkage line loss comprehensive intelligent diagnosis system according to claim 1, characterized in that, The distributed storage and high-performance computing module comprises: A distributed storage unit is configured to store multi-source data based on HBase or Hive, including bus archives, line archives, high-voltage user archives, transformer area archives, gateway lists, line gateways, daily current curve data of special public variable users and low-voltage users, daily voltage curve data of line gateways, special public variable users and low-voltage users, daily power curve data of line gateways, special public variable users and low-voltage users, power factor data of line gateways, special public variable users and low-voltage users, electric quantity data of line gateways, special public variable users and low-voltage users, and line loss data of line gateways, special public variable users and low-voltage users; A high-performance computing unit is configured to perform real-time parallel processing on multi-source data based on a Spark or Flink framework. 3.The high-low voltage linkage line loss comprehensive intelligent diagnosis system according to claim 2, characterized in that, The data fusion and processing module comprises: A data fusion unit is configured to construct a dynamic archive through system of "bus-line-special public variable user-low-voltage user"; A line loss index construction unit is configured to construct a line loss calculation model and a bus calculation model, and calculate input and output electric quantities and line loss rates of each distribution line; A data preprocessing unit is configured to perform abnormal value processing, missing value filling and data normalization; The line loss index construction unit comprises: A line loss calculation subunit is configured to calculate line loss according to the following formula: Supply quantity = line gateway positive electric quantity + public variable transformer area gateway negative electric quantity + special variable user negative electric quantity + high-voltage office electric quantity + high-voltage distributed energy grid-connected electric quantity; Consumption quantity = special variable user positive electric quantity + public variable transformer area gateway positive electric quantity + high-voltage office electric quantity + line gateway negative electric quantity; Line loss rate = (supply quantity-consumption quantity) / supply quantity × 100%; A bus calculation subunit is configured to calculate bus imbalance rate according to the following formula: Bus imbalance rate = (bus supply quantity-bus consumption quantity) / bus supply quantity × 100%.

4. The high-low voltage linkage line loss comprehensive intelligent diagnostic system according to claim 3, characterized in that, The multi-dimensional feature construction module comprises: A time dimension feature unit is configured to extract load change features at different time scales, including mean value, standard deviation, maximum value, minimum value features, and day load curve fluctuation coefficient of electric quantity and line loss at different periods; A space dimension feature unit is configured to calculate high and low voltage network space conductivity features, including electric quantity similarity coefficient, loss electric quantity similarity coefficient and line loss rate similarity coefficient; An electrical dimension feature unit is configured to calculate voltage level, fluctuation, imbalance degree and phase deviation, current level, fluctuation, imbalance degree, phase deviation and trend correlation, power level, fluctuation and total difference, power factor level and fluctuation index, and voltage, current and power cross features at the same time section.

5. The high-low pressure linkage line loss comprehensive intelligent diagnostic system according to claim 4, characterized in that, The reverse coupling relationship is that when an abnormality occurs at a certain metering gateway, the line loss of the upper level and the line loss of the level where the gateway is located show a kind of complementary and reverse fluctuation correlation characteristics, that is, the line loss on one side increases while the line loss on the other side decreases.

6. The high and low voltage linked line loss comprehensive intelligent diagnosis system according to claim 5, wherein the line loss intelligent diagnosis module comprises: A system state model unit is configured to construct a line system state model based on energy conservation, and calculate metering point error rate. The isolated forest anomaly diagnosis unit is configured to identify abnormal data by constructing an isolated forest anomaly detection model; The model fusion unit is configured to generate high-confidence abnormal clues and secondary clues by integrating output results of the system state model and the isolated forest anomaly detection model; The system state model unit is constructed based on the following formula: ； wherein is the number of days is the power supply amount of the antenna line; is the number of days is the power consumption of the first metering point, represents metering points; is the error rate of the first metering point; is the conductor loss of the first day; is the copper loss of the distribution transformer of the first day, which belongs to variable loss; is the iron loss of the distribution transformer, which is constant loss; is the model residual error; The isolated forest anomaly diagnosis unit realizes anomaly detection by the following steps: Multiple isolated trees are constructed to form an isolated forest; Path lengths of samples in the isolated trees are calculated; Anomaly scores are calculated according to the path lengths to determine abnormal samples; The model fusion unit generates abnormal clues by the following steps: Abnormal samples that are agreed by the system state model and the isolated forest anomaly detection model are extracted as high-confidence clues; Non-agreed abnormal samples are subjected to archive specification verification, new energy user feature filtering and time series continuity verification to generate secondary clues.

7. A line loss comprehensive intelligent diagnosis method based on the system of any one of claims 1-6, characterized in that, The method comprises the following steps: Step S1: Real-time processing of multi-source data by a distributed storage and high-performance computing module; Step S2: Fusion of bus-line-user archive data and calculation of line loss indicators; Step S3: Extraction of time, space and electrical three-dimensional feature indicators; Step S4: High-low voltage linkage analysis based on Pearson correlation coefficients and DTW distances; Step S5: Fusion of the system state model and the isolated forest anomaly detection model to output abnormal diagnosis results.

8. The method of claim 7, wherein, The high-low voltage linkage analysis in Step S4 specifically comprises: When the Pearson correlation coefficient r of the line loss and the line loss of the public variable transformer area is <-0.7 and the DTW distance is less than a preset threshold, it is determined that the public variable metering point is abnormal; When the Pearson correlation coefficient |r| of the line loss and the electric quantity of the special variable user is >0.8, it is determined that the special variable user is abnormal.

Images