A method and system for tracing and locating abnormal low-voltage branch line losses
By constructing a radial tree network model and a constrained optimization model for low-voltage distribution transformer substations, and combining Kirchhoff's current law and time-varying impedance calculation, the problem of untraceable line loss anomalies in low-voltage distribution transformer substations was solved, and accurate location of branch line losses and identification of fault types were achieved.
Patent Information
- Application Number
- CN202611328183.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-31
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies cannot pinpoint the specific branch segment where abnormal line loss occurs in low-voltage distribution transformer areas, nor can they trace the source through simple measurement comparisons.
By acquiring the topology files, line ledger parameters, transformer head-end measurement data, and user end measurement data of low-voltage distribution substations, a radial tree network model is constructed. The time-varying impedance is calculated, and Kirchhoff's current law and constrained optimization model are used to solve the problem. Abnormal branch segments are identified, and fault type and spatial location are traced and located.
It enables accurate location of abnormal branch segments with line loss without the need for additional measurement hardware, reducing false alarm rate, improving operation and maintenance efficiency, and is economically feasible and physically interpretable.
Smart Images

Figure CN123066485A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system monitoring technology, and in particular to a method and system for tracing and locating abnormal low-voltage branch line losses. Background Technology
[0002] Low-voltage distribution substations are the last link in the power system that directly faces end users, and their line loss management level is directly related to the economic benefits of power supply companies and the operating efficiency of the power grid.
[0003] Currently, power supply companies generally install smart converged terminals (TTUs) at the low-voltage outgoing lines of distribution transformers and smart meters at each user metering point. These systems automatically collect measurement data from the transformer's head-end and user's end-end every 15 minutes using an electricity consumption information collection system. Based on this measurement data, the line loss management system can calculate the total line loss rate at the transformer substation level, which is the percentage of the difference between the total power at the transformer's head-end and the total power of all users relative to the total power at the transformer's head-end. When the line loss rate of a substation exceeds a preset threshold, the system generates a line loss anomaly alarm, prompting maintenance personnel to investigate the substation.
[0004] The existing line loss anomalies cannot be pinpointed to specific branches. Low-voltage distribution transformer areas have a radial, tree-like network structure, branching off from the transformers through main trunk lines to individual users, typically involving multiple branch lines. When a line loss anomaly occurs in the entire transformer area, maintenance personnel only know that the anomaly exists, but cannot determine which branch line the anomaly occurs on. Because branch lines typically do not have direct electrical measurement devices such as current transformers or power sensors installed, it is impossible to directly obtain current and power data for each branch line. Therefore, it is impossible to locate the abnormal branch through simple measurement comparisons. Summary of the Invention
[0005] The purpose of this invention is to provide a method and system for tracing and locating abnormal low-voltage branch line losses, so as to solve the technical problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for tracing and locating abnormal low-voltage branch line losses, applied to low-voltage distribution transformer areas, includes: The topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data and ambient temperature data of the low-voltage distribution transformer area are obtained, and a radial tree network model is constructed based on the topology file to divide the low-voltage distribution transformer area into multiple branch segments. The time-varying impedance of each branch line segment is obtained based on the line ledger parameters and ambient temperature data, and the current distribution relationship of each branch line segment is constructed according to Kirchhoff's current law. A constrained optimization model is established based on the measurement data at the head end of the distribution transformer and the measurement data at each user end; the constrained optimization model is solved to obtain the estimated actual line loss of each branch segment; Obtain preset judgment conditions, and determine abnormal branch segments based on the preset judgment conditions and the estimated actual line loss of each branch segment; Based on the estimated actual line loss of the abnormal branch segment, the measurement data at the head end of the distribution transformer and the measurement data at the user end, multi-dimensional abnormal physical features are extracted, and fault types and abnormal evidence chains are generated based on the multi-dimensional abnormal physical features. The spatial location of the fault on the abnormal branch segment is obtained based on the estimated actual line loss of the abnormal branch segment; The source tracing and location results are generated based on the abnormal branch segments, fault types, spatial locations, and abnormal evidence chains.
[0007] Preferably, the step of obtaining the time-varying impedance of each branch segment based on the line ledger parameters and ambient temperature data includes: The basic impedance value of each branch segment at the reference temperature is obtained based on the line ledger parameters. Based on the ambient temperature data and the branch current estimate obtained from the previous iteration, the current conductor temperature is calculated according to the linear superposition relationship between conductor temperature and the ambient temperature and the square of the branch current. The base impedance value is corrected based on the current conductor temperature and temperature coefficient of resistance to obtain the branch impedance value at the current iteration number; Based on the branch impedance value, the constraint optimization model is solved again to obtain the updated branch current estimate. The updated branch current estimate is used to determine whether the convergence criterion is met. The convergence criterion is that the absolute value of the change in the current conductor temperature between two adjacent iterations is less than a first preset threshold and the relative change in the branch impedance value between two adjacent iterations is less than a second preset threshold. If the convergence criterion is met, the branch impedance value at the current iteration number is output as the time-varying impedance; if the convergence criterion is not met and the iteration number does not exceed the preset maximum iteration number, the branch current estimate obtained based on the ambient temperature data and the previous iteration is returned.
[0008] Preferably, the step of establishing a constrained optimization model based on the measurement data at the head end of the distribution transformer and the measurement data at each user end includes: Based on the active power, reactive power and voltage measured at each user terminal, the current estimate of each user is calculated according to the inverse relationship that current equals the ratio of apparent power to voltage. Then, based on the radial tree network model, the current estimate of each user is summarized according to the branch segments to obtain the soft estimate of the branch current of each branch segment. The measured bus loss of the distribution area is obtained by subtracting the total power of all users from the total power of the distribution transformer head end measurement data in the distribution transformer head end measurement data; Based on the topology of the radial tree network model, an equality constraint is generated to ensure that the sum of the branch currents of each branch segment equals the total current at the head of the transformer. The soft estimate of the branch current and the measured bus loss of the transformer area are used as the measurement inputs of the constraint optimization model, and the equality constraint is used as the structural constraint of the constraint optimization model.
[0009] Preferably, the step of obtaining preset judgment conditions and determining abnormal branch segments based on the preset judgment conditions and the estimated actual line loss of each branch segment includes: Based on the minimum active power resolution of the intelligent fusion terminal that collects the measurement data at the head end of the distribution transformer and the smart meter that collects the measurement data at the user end, an absolute threshold is determined; the difference between the actual line loss estimate and the corresponding theoretical line loss value of each branch segment is taken as the line loss deviation value, where the theoretical line loss value is the line loss value calculated according to Joule's law based on the time-varying impedance and the corresponding soft estimate of the branch current; it is then determined whether the absolute value of the line loss deviation value is greater than the absolute threshold. If it is, the absolute deviation determination result is satisfied; otherwise, the absolute deviation determination result is not satisfied. Based on the comprehensive uncertainty of the theoretical line loss value obtained by error propagation calculation of the manufacturing tolerance of conductor cross-sectional area and the measurement error of conductor length in the line ledger parameters, a relative threshold is determined; it is then determined whether the ratio of the absolute value of the line loss deviation value to the theoretical line loss value is greater than the relative threshold. If so, the relative deviation determination result is determined to be satisfied; otherwise, it is determined to be unsatisfactory. The standard uncertainty of the estimated actual line loss obtained by error propagation from the measurement error covariance matrix during the solution of the constrained optimization model is multiplied by a preset confidence factor to determine the statistical threshold; it is then determined whether the absolute value of the line loss deviation is greater than the statistical threshold. If so, the statistical significance determination result is deemed to be satisfied; otherwise, it is deemed not to be satisfied. When the absolute deviation determination result, the relative deviation determination result, and the statistical significance determination result of the same branch segment are all satisfied, the branch segment is determined to be an abnormal branch segment.
[0010] Preferably, the step of generating the fault type based on the multidimensional abnormal physical characteristics includes: Obtain the target value or value range of each preset fault type in each feature dimension, and generate the feature signature vector corresponding to each preset fault type. The fault physical feature signature library is composed of multiple feature signature vectors. Based on the estimated actual line loss of the abnormal branch segment, the measurement data at the head of the distribution transformer, and the measurement data at the user end, the multidimensional abnormal physical features are extracted. Based on the standard deviation of each feature dimension under normal conditions, the matching weight of each feature dimension is determined according to the relationship that the weight is equal to the inverse square of the standard deviation and then normalized. Based on the matching weight, the weighted distance between the multidimensional abnormal physical feature and each feature signature vector is calculated, and the preset fault type corresponding to the feature signature vector with the smallest weighted distance is determined as the candidate fault type. Obtain the residual threshold and deviation threshold. Based on the comparison result of Kirchhoff's Current Law residual and residual threshold in the multidimensional abnormal physical characteristics, determine whether to exclude electricity theft signatures. Based on the comparison result of neutral current deviation ratio and deviation threshold, determine whether to exclude leakage signatures. Filter the candidate fault types to obtain the fault types.
[0011] Preferably, the step of obtaining the spatial location of the fault on the abnormal branch segment based on the estimated actual line loss of the abnormal branch segment includes: The impedance estimate of the abnormal branch segment is obtained based on the solution results of the constrained optimization model. The theoretical impedance value at the operating temperature is calculated based on the ambient temperature data and the estimated branch current of the abnormal branch segment. The theoretical impedance value is then subtracted from the estimated impedance value to obtain the impedance deviation after deducting the temperature effect. Based on the impedance deviation and the resistance per unit length of the abnormal branch segment, calculate the equivalent distance from the fault point to the beginning of the branch. The deviation time trend is obtained based on the multidimensional abnormal physical characteristics, and the equivalent distance is corrected based on the deviation time trend: when the time trend is a gradual trend, the nearest joint position record is found in the line ledger parameters based on the equivalent distance, and the position range corresponding to the joint position record is used as the spatial position; when the time trend is a sudden trend, the equivalent distance is directly used as the spatial position.
[0012] This invention discloses a low-voltage branch line loss anomaly tracing and location system, comprising: The partitioning module is used to acquire the topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data and ambient temperature data of the low-voltage distribution transformer area, and construct a radial tree network model based on the topology file to divide the low-voltage distribution transformer area into multiple branch segments. The allocation module is used to obtain the time-varying impedance of each branch line segment based on the line ledger parameters and ambient temperature data, and to construct the current distribution relationship of each branch line segment according to Kirchhoff's current law. The actual line loss estimation module is used to establish a constrained optimization model based on the measurement data at the head end of the distribution transformer and the measurement data at the end of each user; and to solve the constrained optimization model to obtain the actual line loss estimation value of each branch segment. An abnormal branch determination module is used to obtain preset determination conditions and determine abnormal branch segments based on the preset determination conditions and the estimated actual line loss of each branch segment. The fault type generation module is used to extract multi-dimensional abnormal physical features based on the actual line loss estimate of the abnormal branch segment, the measurement data at the head end of the distribution transformer and the measurement data at the user end, and to generate fault types and abnormal evidence chains based on the multi-dimensional abnormal physical features. The spatial location acquisition module is used to obtain the spatial location of the fault on the abnormal branch segment based on the estimated actual line loss of the abnormal branch segment; The result generation module is used to generate source tracing and location results based on the abnormal branch segments, fault types, spatial locations, and abnormal evidence chains.
[0013] Preferably, the allocation module includes: The basic impedance value acquisition unit is used to acquire the basic impedance value of each branch line segment at the reference temperature based on the line ledger parameters. The current conductor temperature acquisition unit is used to calculate the current conductor temperature based on the ambient temperature data and the estimated branch current obtained from the previous iteration, according to the linear superposition relationship between conductor temperature and the ambient temperature and the square of the branch current. The branch impedance value acquisition unit is used to correct the basic impedance value based on the current conductor temperature and the temperature coefficient of resistance to obtain the branch impedance value under the current iteration number. The branch current estimation unit is used to re-execute the solution of the constraint optimization model based on the branch impedance value to obtain the updated branch current estimation value. The judgment unit is used to determine whether the convergence criterion is met based on the updated branch current estimate, wherein the convergence criterion is that the absolute value of the change in the current conductor temperature between two adjacent iterations is less than a first preset threshold and the relative change in the branch impedance value between two adjacent iterations is less than a second preset threshold. If the convergence criterion is met, the branch impedance value at the current iteration number is output as the time-varying impedance; if the convergence criterion is not met and the iteration number does not exceed the preset maximum iteration number, the branch current estimate obtained based on the ambient temperature data and the previous iteration is returned.
[0014] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for tracing and locating abnormal low-voltage branch line losses.
[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for tracing and locating anomalies in low-voltage branch lines.
[0016] The beneficial effects of this application are as follows: This invention constructs virtual measurements based on transformer head-end measurement data and user end-end measurement data under the condition that there are no direct electrical measurements on branch lines. It establishes a constraint optimization model for solving the model, using the non-negativity of branch line losses and the sum of these losses equaling the measured total line loss in the transformer area as a hard constraint, and the sparsity of the branch line loss deviation vector as a regularization constraint. This allows for the decomposition of the transformer area's total line loss to each branch line segment using only existing measurement equipment, achieving branch-level line loss anomaly localization without the need for any additional measurement hardware, thus possessing economic feasibility for large-scale promotion. Furthermore, by deriving the theoretical manifestations of various faults in each characteristic dimension based on Kirchhoff's current law, Ohm's law, and the Joule heating effect, a fault physical feature signature library is constructed. Multidimensional abnormal physical features are matched with this library to determine the fault type. Simultaneously, an abnormal evidence chain recording the physical law basis corresponding to each feature component is generated, making the diagnostic conclusions physically interpretable and verifiable. The traceability feature overcomes the shortcomings of existing data-driven methods that rely on a large number of labeled samples and lack physical interpretability. By calculating the impedance deviation after deducting the temperature effect and dividing it by the resistance per unit length to obtain the equivalent distance, and then combining the deviation time trend and the joint location records in the line ledger parameters for correction, the spatial location of the fault is estimated, narrowing the investigation scope from the entire branch line segment to a specific joint or distance range. By coupling and iteratively calculating the current conductor temperature based on the ambient temperature data and the estimated branch current value and correcting it to obtain the time-varying impedance, it accurately reflects the true resistance level of the conductor under actual load conditions, avoiding misjudgments caused by the systematic underestimation of theoretical line loss values under high temperature and high load conditions in summer. By setting absolute thresholds, relative thresholds, and statistical thresholds from three dimensions—measurement resolution, model uncertainty, and statistical significance—for joint judgment, false deviations caused by non-fault factors are effectively eliminated, significantly reducing the false alarm rate. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.
[0018] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.
[0019] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.
[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0021] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0022] like Figure 1 As shown, this application provides a method for tracing and locating abnormal low-voltage branch line losses, applied to low-voltage distribution transformer areas, including: S1. Obtain the topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data and ambient temperature data of the low-voltage distribution substation, and construct a radial tree network model based on the topology file to divide the low-voltage distribution substation into several branch segments. S2. Calculate the temperature-corrected time-varying impedance for each branch line segment based on the line ledger parameters and ambient temperature data, and construct the current distribution relationship for each branch line segment according to Kirchhoff's current law. S3. Under the condition that there is no direct electrical measurement of the branch line segment, construct virtual measurement based on the measurement data of the transformer head end and the measurement data of each user end, and establish a constraint optimization model based on the virtual measurement. The constraint optimization model takes the non-negative line loss of each branch and the sum of them equal to the actual line loss of the transformer area as hard constraint, and the sparsity of the deviation vector of each branch line loss as regularization constraint. Solve the constraint optimization model to obtain the estimated value of the actual line loss of each branch line segment.
[0023] S4. Based on the deviation between the actual line loss estimate and the theoretical line loss, when the deviation meets the preset judgment condition, determine the abnormal branch segment; S5. Extract multidimensional abnormal physical features based on the estimated actual line loss of the abnormal branch segment, the measurement data at the head end of the distribution transformer and the measurement data at the user end. Match the multidimensional abnormal physical features with the preset fault physical feature signature library to determine the fault type. Generate an abnormal evidence chain containing the physical laws corresponding to each feature component based on the matching results. S6. Based on the impedance deviation between the actual line loss estimate and the theoretical line loss value of the abnormal branch segment, estimate the spatial location of the fault on the abnormal branch segment; S7. Output the source location result including the abnormal branch segment, the fault type, the spatial location, and the abnormal evidence chain.
[0024] As described in steps S1-S7 above, this application first obtains the topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data, and ambient temperature data for the low-voltage distribution substation. The topology file refers to the electrical connection relationship file between the transformer and each user in the low-voltage distribution substation, including the transformer outgoing circuit number, the head-end node numbers of each branch line segment, the branch line segment number connected to each user, and phase affiliation information. This topology file is usually stored in the power supply company's marketing system or distribution management system. The line ledger parameters refer to the physical parameter records of each branch line segment, including conductor type, conductor cross-sectional area, conductor length, conductor material (aluminum or copper), resistance per unit length, and joint location records. These line ledger parameters are usually stored in the equipment ledger management system. The transformer head-end measurement data... The data refers to the electrical measurement data collected by the smart converged terminal (TTU) installed at the low-voltage side outgoing line of the distribution transformer, including three-phase active power, three-phase reactive power, three-phase voltage, three-phase current, and neutral current, etc., with a collection cycle of generally 15 minutes; the user-end measurement data refers to the electrical measurement data collected by the smart meters installed at each user metering point, including active power, reactive power, voltage, current, etc., with a collection cycle of generally 15 minutes; the ambient temperature data refers to the ambient temperature of the area where the transformer substation is located, which can be obtained through temperature sensors installed at the distribution transformer or through the meteorological service interface.
[0025] By acquiring the aforementioned multi-source data, a comprehensive understanding of the network structure, line physical parameters, electrical measurements at the beginning and end points, and environmental conditions of the low-voltage distribution transformer substation can be obtained. A radial tree network model is constructed based on the topology archive. This model is a directed tree diagram with the transformer as the root node, each branch segment as a tree branch, and each user as a leaf node. The network structure of a low-voltage distribution transformer substation naturally exhibits a radial distribution, meaning that branches sequentially from the transformer outgoing line to each user, without loops. Therefore, a tree model can accurately represent its topological characteristics. The low-voltage distribution transformer substation is divided into several branch segments. Each branch segment is a line segment between two adjacent nodes in the tree model. For example, a substation may have 5 branch segments, denoted as B1, B2, B3, B4, and B5. B1 is the main line segment from the transformer outgoing line to the first branch node, and B2 to B5 are the branch segments from each branch node to the end user.
[0026] Because the conductor resistance of low-voltage lines varies with temperature, using a fixed resistance value from the ledger for line loss calculation will result in significant errors under high summer temperatures or low winter temperatures. Therefore, it is necessary to calculate the temperature-corrected time-varying impedance for each branch segment based on the line ledger parameters and ambient temperature data, so that the impedance value can reflect the true resistance level under the current operating conditions. Simultaneously, the current distribution relationship of each branch segment is constructed according to Kirchhoff's Current Law, that is, at each branch node in the tree network, the current flowing into that node is equal to the sum of the currents flowing out of that node from all branches. This current distribution relationship forms the structural basis for subsequently establishing the constrained optimization model.
[0027] In low-voltage distribution substations, direct electrical measurement devices are typically not installed on branch lines, meaning it's impossible to directly obtain electrical quantities such as current and power for each branch line. Measurement data only exists at the transformer head and each user terminal. Under these conditions of no direct electrical measurement on branch lines, this application constructs virtual measurements based on the transformer head and user terminal measurement data. Virtual measurements refer to equivalent measurement information not directly acquired by physical sensors, but calculated from existing measurement data and physical relationships. A constrained optimization model is established based on the virtual measurements. Under the premise of satisfying the physical conservation law (i.e., the line loss of each branch is non-negative and its sum equals the measured line loss of the entire substation), the model utilizes the spatial sparsity of line loss anomalies (i.e., at any given time, the vast majority of branch lines are in normal condition, with only a few showing anomalies). Using the sparsity of the line loss deviation vector of each branch as a regularization constraint, the most probable line loss value for each branch line is mathematically solved. After solving the constrained optimization model, the estimated actual line loss of each branch segment is obtained. This estimated actual line loss reflects the true line loss level of each branch segment under the current operating conditions.
[0028] The estimated actual line loss is compared with the theoretical line loss, which is an ideal line loss calculated based on the conductor resistance and current in the line log parameters. When the deviation between the two meets the preset judgment criteria, it indicates that the actual line loss of the branch segment has significantly deviated from the normal level, and it is identified as an abnormal branch segment. The design of the preset judgment criteria needs to balance sensitivity and reliability, ensuring that it can detect real line loss anomalies while avoiding misjudgments caused by measurement noise or model errors.
[0029] After identifying the abnormal branch segment, it is necessary to further trace the physical causes of the abnormal line loss. Based on the estimated actual line loss of the abnormal branch segment, the measurement data at the transformer's head end, and the user's end measurement data, multi-dimensional abnormal physical features are extracted. These multi-dimensional abnormal physical features are a set of features characterizing the abnormal phenomenon from multiple physical dimensions, such as the magnitude of the line loss deviation, the correlation between the deviation and the square of the current, the correlation between the deviation and voltage, the correlation between the deviation and temperature, whether the neutral current is abnormal, whether Kirchhoff's current law is balanced, the temporal trend of the deviation, and the time-period distribution characteristics of the deviation. Different types of faults (such as electricity theft, leakage, poor contact, and distributed photovoltaic backfeeding) exhibit different physical laws in the above feature dimensions. Therefore, a fault physical feature signature library can be pre-constructed based on physical laws. By matching the multi-dimensional abnormal physical features with the fault physical feature signature library, the most likely fault type can be determined. Meanwhile, since each feature component in the fault physical feature signature library is derived from explicit physical laws, the matching results naturally come with a physically interpretable chain of evidence for anomalies. This chain of evidence records the physical law basis corresponding to each feature component, making the diagnostic conclusions traceable and verifiable.
[0030] After determining the fault type, it is also necessary to estimate the spatial location of the fault. Based on the impedance deviation between the actual estimated line loss and the theoretical line loss of the abnormal branch segment, which reflects the additional resistance effect caused by the fault, dividing it by the resistance per unit length can estimate the distance from the fault point to the beginning of the branch segment, thereby estimating the spatial location of the fault on the abnormal branch segment.
[0031] Finally, the output includes the source location results of the abnormal branch segment, fault type, spatial location, and abnormal evidence chain. Maintenance personnel can directly go to the corresponding branch segment and spatial location for on-site investigation based on these results, eliminating the need for segment-by-segment line inspection and significantly improving maintenance efficiency. It should be noted that the entire calculation process in this application can be implemented in software on the power supply company's line loss management master station or edge computing gateway, without the need to install any additional measuring equipment on the branch segment. Branch-level line loss anomaly source location can be completed using only existing transformer head-end measurement data and user-end measurement data.
[0032] In one embodiment, step S2, which calculates the temperature-corrected time-varying impedance for each branch segment based on the line ledger parameters and ambient temperature data, includes: S21. Obtain the basic impedance value of each branch line segment at the reference temperature according to the line ledger parameters; S22. Based on the ambient temperature data and the estimated branch current obtained from the previous iteration, calculate the current conductor temperature according to the linear superposition relationship between conductor temperature and the square of ambient temperature and branch current. S23. Correct the base impedance value based on the current conductor temperature and temperature coefficient of resistance to obtain the branch impedance value under the current iteration number; S24. Re-execute the solution of the constraint optimization model based on the branch impedance value to obtain the updated branch current estimate; S25. Determine whether the convergence criterion is met based on the updated branch current estimate, wherein the convergence criterion is that the absolute value of the change in the current conductor temperature between two adjacent iterations is less than a first preset threshold and the relative change in the branch impedance value between two adjacent iterations is less than a second preset threshold. S26. If the convergence criterion is met, the branch impedance value at the current iteration number is used as the time-varying impedance output; if the convergence criterion is not met and the iteration number does not exceed the preset maximum iteration number, return to step S22 to continue iterating.
[0033] As described in steps S21-S26 above, when calculating the temperature-corrected time-varying impedance, the basic impedance value of each branch segment at the reference temperature is first obtained based on the line ledger parameters. Specifically, the line ledger parameters record the conductor type, conductor cross-sectional area, conductor length, and conductor material of each branch segment. The resistivity of the conductor at the 20℃ reference temperature is determined based on the conductor material. The formula for calculating the basic impedance value is as follows: ; Among them, R 20 Here, L is the basic impedance value, S is the conductor length, and ρ is the conductor cross-sectional area. 20 Determine the resistivity of the conductor material at a reference temperature of 20℃. For example, a branch line segment uses JKLYJ-1×70 type aluminum core overhead insulated conductor, with a conductor length L of 120 meters and a conductor cross-sectional area S of 70 mm². 2 Then its basic impedance value R 20 =0.0283×120 / 70≈0.0485Ω.
[0034] Since the actual operating temperature of the conductor is not a constant 20°C, but is affected by both the ambient temperature and the load current, it is necessary to calculate the current conductor temperature. This embodiment uses a coupled iterative method for temperature correction. The conductor temperature depends on the load current (the larger the current, the stronger the Joule heating effect, and the higher the conductor temperature), and the conductor temperature, in turn, affects the conductor resistance (the higher the temperature, the greater the resistance). The change in resistance also affects the current distribution. Therefore, there is a bidirectional coupling relationship between temperature and current, which needs to be solved iteratively.
[0035] Specifically, in the first iteration, since there is no branch current estimate yet, the soft estimate of the branch current from the virtual measurement can be used as the initial value. In the k-th iteration, based on the ambient temperature T from the ambient temperature data and the branch current estimate I obtained from the previous iteration, the current conductor temperature is calculated according to the following formula: ; Where T represents the current conductor temperature in the k-th iteration, in °C; T represents the ambient temperature, obtained from ambient temperature data, in °C; β represents the proportionality coefficient between the square of the current and the temperature rise, which physically represents the steady-state temperature rise generated on the conductor per unit square of current. This coefficient is determined according to Appendix A of GB / T44768-2024 based on the conductor material, cross-sectional area, and laying method (overhead or cable). For example, for JKLYJ-1×70 type overhead aluminum conductor, the typical value of β is approximately 0.003 °C / A. 2 I represents the branch current estimate obtained in the (k-1)th iteration, in A, and is initialized by the soft estimate of the branch current in the first iteration.
[0036] After obtaining the current conductor temperature, the base impedance value is corrected based on the current conductor temperature and the temperature coefficient of resistance to obtain the branch impedance value for the current iteration number. The correction formula is as follows: ; Among them, R (k) R represents the branch impedance value in the k-th iteration, in Ω. 20 The base impedance value at a reference temperature of 20℃ is calculated from the line ledger parameters and is in Ω; α represents the temperature coefficient of resistance, which is a known constant. This represents the current conductor temperature in the k-th iteration, calculated using the previous formula, and is expressed in °C.
[0037] The constrained optimization model is re-solved based on the corrected branch impedance value. Since the impedance value has changed, the current distribution of each branch segment is updated accordingly, resulting in an updated branch current estimate. Then, the convergence criterion is checked. The convergence criterion includes two conditions: first, the absolute value of the change in current conductor temperature between two adjacent iterations is less than a first preset threshold (e.g., 0.1℃); second, the relative change in branch impedance between two adjacent iterations is less than a second preset threshold (e.g., 0.1%). Both conditions must be met simultaneously for the iteration to be considered converged. If the convergence criterion is met, the branch impedance value at the current iteration number is used as the final time-varying impedance output. If the convergence criterion is not met and the number of iterations has not exceeded the preset maximum number of iterations (e.g., 20), the process returns to step S22, recalculates the current conductor temperature using the updated branch current estimate, and continues to the next iteration. If the iteration count has reached the preset maximum number of iterations and convergence has not yet occurred, the current iteration result is used as the time-varying impedance output, and an alarm message is generated to alert maintenance personnel.
[0038] It should be noted that using a coupled iterative method to calculate time-varying impedance, compared to the traditional "one-time correction based on fixed ambient temperature" method, can more accurately reflect the true resistance level of the conductor under actual load conditions. Especially in high-temperature summer conditions with high load rates, the actual conductor temperature may be much higher than the ambient temperature. If the temperature rise caused by the current heating effect is ignored, the theoretical line loss value will be systematically underestimated, leading to normal branch segments being misjudged as abnormal branch segments.
[0039] In one embodiment, step S3, which establishes a constrained optimization model based on the distribution transformer head-end measurement data and the user end measurement data, includes: S31. Based on the active power, reactive power and voltage in the measurement data of each user terminal, calculate the current estimate of each user according to the inverse relationship that current equals the ratio of apparent power to voltage, and summarize the current estimate of each user according to the branch segment according to the radial tree network model to obtain the soft estimate of the branch current of each branch segment. S32. Subtract the total power of all users from the total power of the transformer head end measurement data in the transformer head end measurement data to obtain the actual bus loss of the transformer area. S33. Based on the topology of the radial tree network model and Kirchhoff's current law, generate an equality constraint that the sum of the branch currents of each branch segment equals the total current at the head of the transformer. S34. The soft estimate of the branch current and the measured bus loss of the transformer area are used as the measurement inputs of the constraint optimization model, and the equality constraint is used as the structural constraint of the constraint optimization model.
[0040] As described in steps S31-S34 above, in low-voltage distribution substations, direct electrical measurement devices such as current transformers or power sensors are typically not installed on branch lines. Instead, smart fusion terminals (TTUs) are installed only at the transformer head end, and smart meters are installed at each user metering point. Therefore, it is impossible to directly obtain the current and power data for each branch line; virtual measurements need to be constructed using existing measurement data.
[0041] Specifically, in step S31, based on the active power ρ, reactive power Q, and voltage V from the measurement data at each user's terminal, the estimated current value for each user is calculated according to the inverse relationship that current equals the ratio of apparent power to voltage. The physical basis of this inverse relationship is: in a single-phase AC circuit, apparent power equals the product of voltage and current; therefore, current equals apparent power divided by voltage. Apparent power is equal to the square root of the sum of the squares of active power and reactive power. The calculation formula is: ; Among them, I est,j P represents the estimated current for the j-th user, in amperes (A). j Q represents the active power of the j-th user, obtained from user-end measurement data, in W; j This represents the reactive power of the j-th user, obtained from user-end measurement data, in var (V). j This represents the voltage of the j-th user, obtained from user terminal measurement data, in volts (V). For example, if a user has an active power of 2200W, a reactive power of 500var, and a voltage of 220V, then its estimated current is √(2200V). 2 +500 2 ) / 220≈10.25A.
[0042] After obtaining the current estimates for each user, the estimated current values for each user are summarized according to their respective branch segments based on the radial tree network model. Specifically, the radial tree network model records the branch segment number to which each user is connected. The current estimates of all users connected to the same branch segment are summed to obtain the soft estimate of the branch current for that branch segment. For example, if branch segment B3 has five users (users 12, 13, 14, 15, and 16) connected to it, with estimated current values of 9.5A, 8.2A, 7.8A, 5.5A, and 4.2A respectively, then the soft estimate of the branch current for branch segment B3 is 9.5 + 8.2 + 7.8 + 5.5 + 4.2 = 35.2A. This is called a "soft estimate" because it is derived from the end-point measurement data and contains certain measurement and model approximation errors; it is not a "hard measurement" directly collected by sensors installed on the branch segment.
[0043] In step S32, the measured line loss of the transformer substation is obtained by subtracting the total power of all users from the measured power of the transformer substation's head-end data. The physical basis of this calculation is the law of conservation of energy: the total power output by the transformer equals the total power consumed by all users plus the total power lost by all lines within the transformer substation. Therefore, the line loss of the transformer substation equals the total power at the transformer head-end minus the total power of the users. The calculation formula is: ; in, This indicates the measured bus loss in the transformer area, in W; P TTU This represents the total power at the transformer's head end, obtained from measurement data at the transformer's head end, and is expressed in watts (W); N u The total number of users is obtained from the topology file; P j This represents the active power of the j-th user, obtained from the user's terminal measurement data, in watts (W). For example, if the total power at the head end of a transformer substation is 50,000W, and there are 20 users with a total active power of 48,500W, then the measured bus loss for the substation is 50,000 - 48,500 = 1,500W. This measured bus loss is a "hard measurement" value because it only depends on the difference between the power measurements at the head end and the terminal end, without involving any model assumptions or parameter estimations, and therefore has high reliability.
[0044] In step S33, based on the topology of the radial tree network model and Kirchhoff's current law, an equality constraint is generated whereby the sum of the branch currents of each branch segment equals the total current at the transformer head. The physical basis of this equality constraint is: in the radial tree network, the transformer head is the only power injection point, and the current in all branch segments ultimately originates from the transformer head. Therefore, the sum of the currents in all branch segments on the same phase should equal the total current at the transformer head on that phase. The calculation formula is: ; Where, N b The total number of branch segments is determined by the radial tree network model; I i,φ Let represent the branch current of the i-th branch segment in phase φ, which is the variable to be solved in the constrained optimization model, and its unit is A; This represents the total current at the transformer's head end in phase φ, obtained from measurement data at the transformer's head end, with units of A; φ represents the values of phases A, B, and C. This equation constraint is established separately for each phase, forming a total of three equation constraints.
[0045] In step S34, the soft estimate of the branch current and the measured line loss of the transformer area obtained above are used as the measurement inputs of the constrained optimization model, and the equality constraints are used as the structural constraints of the constrained optimization model. The soft estimate of the branch current provides prior estimation information of the current in each branch segment, the measured line loss of the transformer area provides hard constraint information on the total line loss, and the equality constraints ensure the physical self-consistency of the solution results. These three together constitute the complete input of the constrained optimization model.
[0046] In one embodiment, the process of establishing the constrained optimization model in step S3, which involves establishing the constrained optimization model based on the virtual measurement, includes: S35. Construct a measurement vector based on the virtual measurement, and construct a measurement error covariance matrix based on the accuracy parameters of the measurement device; S36. Construct a theoretical line loss vector based on the theoretical line loss value of each branch segment, and construct a ledger impedance parameter vector based on the line ledger parameters. S37. The weighted sum of squared residuals between the measurement vector and the state variables of the constrained optimization model is used as the data fitting term, the L2 norm of the difference between each branch line loss vector and the theoretical line loss vector is used as the shrinkage regularization term, the L1 norm of the difference between each branch line loss vector and the theoretical line loss vector is used as the sparsity regularization term, and the L2 norm of the difference between the impedance parameter vector and the ledger impedance parameter vector is used as the parameter regularization term. The weighted sum of the data fitting term, shrinkage regularization term, sparsity regularization term and parameter regularization term is used as the objective function. S38. The objective function is solved with the constraint that the line loss of each branch is non-negative and the sum of the line losses of each branch is equal to the measured line loss of the transformer area.
[0047] As described in steps S35-S38 above, establishing a constrained optimization model is a key step in transforming the problem of tracing and locating the source of line loss anomalies into a mathematical optimization problem.
[0048] In step S35, a measurement vector is constructed based on the virtual measurements. The measurement vector is a column vector formed by arranging all virtual measurements in a fixed order, specifically including: the soft estimate of the branch current for each branch segment (arranged by branch segment number) and the measured bus loss of the transformer area. For example, if a transformer area has 5 branch segments, the measurement vector z is a 6-dimensional column vector, with the first 5 components being the soft estimate of the branch current for the 5 branch segments, and the 6th component being the measured bus loss of the transformer area. Simultaneously, a measurement error covariance matrix R is constructed based on the accuracy parameters of the measurement devices. The accuracy parameters of the measurement devices include: the active power measurement accuracy of the smart fusion terminal (TTU) (typically 0.5 class), and the active power measurement accuracy of the smart meter (typically class 1 or 2), etc. Based on the accuracy level and range of each measurement channel, the standard uncertainty of each measurement channel is calculated, and then a diagonal measurement error covariance matrix R is constructed, with its diagonal elements representing the variance of each measurement channel.
[0049] In step S36, a theoretical line loss vector x is constructed based on the theoretical line loss values of each branch segment. The theoretical line loss value is the ideal line loss value calculated according to Joule's law based on the time-varying impedance and the current branch current estimate, i.e., x = I. 2 ×R, where I is the estimated branch current of the i-th branch segment, and R is the branch impedance value of the i-th branch segment at the current iteration number. The theoretical line loss vector is obtained by arranging the theoretical line loss values of all branch segments by number. Simultaneously, a ledger impedance parameter vector θ is constructed based on the line ledger parameters; this vector is composed of the basic impedance values of each branch segment arranged by number.
[0050] In step S37, the objective function is constructed. The objective function consists of four weighted terms: The first term is the data fitting term, which represents the weighted sum of squared residuals between the measurement vector z and the measurement function h(x,θ). Its physical meaning is to require the output of the optimization model to be as close as possible to the actual measurement data. The weights are determined by the inverse matrix of the measurement error covariance matrix R. The higher the accuracy of the measurement channel, the greater the weight.
[0051] The second term is the contraction regularization term, which represents the squared L2 norm of the difference between the line loss vector x of each branch and the theoretical line loss vector x. Its physical meaning is to require that the solution result does not deviate too far from the theoretical expectation, thus preventing overfitting. The strength parameter λ of this regularization term is determined based on the reciprocal of the measurement noise power. The greater the measurement noise, the larger the value of λ, and the stronger the constraint on the solution result.
[0052] The third term is the sparsity regularization term, representing the norm of the difference between the branch line loss vector x and the theoretical line loss vector x. Its physical meaning is based on the prior knowledge that line loss anomalies are spatially sparsity, meaning that at any given time, the vast majority of branch segments are in a normal state (their actual line loss is close to the theoretical line loss), with only a few branch segments exhibiting significant deviations. Norm regularization can cause the deviation of most components in the solution result to approach zero, retaining only a few non-zero deviation components, thereby achieving automatic screening of abnormal branch segments. The strength parameter λ of this regularization term is determined based on the prior knowledge that the number of abnormal branch segments does not exceed the floor value of the total number of branch segments divided by five, i.e., assuming that the number of abnormal branch segments at any given time does not exceed... ; The fourth term is the parameter regularization term, which represents the squared L2 norm of the difference between the impedance parameter vector θ and the ledger impedance parameter vector θ. Its physical meaning is that it allows the impedance parameter to be slightly adjusted near the ledger value to adapt to actual operating conditions, but does not allow excessive deviation to prevent parameter estimation divergence. The strength parameter λ of this regularization term is determined based on the reliability of the ledger parameters. When the ledger parameters have been recently verified in the field and have high reliability, λ takes a larger value; when the ledger parameters are old and have not been verified, λ takes a smaller value.
[0053] The complete expression of the objective function is: ; Where x represents the line loss vector of each branch, which is the decision variable of the constrained optimization model; θ represents the impedance parameter vector, which is the decision variable of the constrained optimization model; z represents the measurement vector, which is constructed from virtual measurements; h(·) represents the measurement function, which is determined by the physical equation of the radial tree network model; R represents the measurement error covariance matrix; x represents the theoretical line loss vector; θ represents the ledger impedance parameter vector; λ1, λ2, and λ3 represent the strength parameters of the sparse regularization term, the contraction regularization term, and the parameter regularization term, respectively, where λ1 (sparse regularization): controls the sparsity of abnormal branch screening. First, the least squares solution without regularization is calculated, and the maximum effective value λ is determined accordingly. 1,max And then Binary search within the interval makes the number of branches with non-zero bias close to Typical value: N b When there are 5 to 20 items, take 10³ to 10¹, N b When there are more than 20 items, take 10. 4 ~10².
[0054] λ2 (Contraction Regularization): Prevents overfitting of the solution to measurement noise. According to the formula: γ, calculate, where, To measure noise power, The theoretical line loss variance is γ∈[0.5,2]. Typical values range from 0.01 to 0.1.
[0055] λ3 (parameter regularization): The extent to which the constraint impedance parameter deviates from the ledger value. Calculated using the formula: ; in It is obtained by error propagation from the conductor cross-sectional area tolerance (±3%) and length error (±5%). Typical values: 50~200 under normal operating conditions, 500~1000 when the ledger is verified and reliable, and 0.1~1.0 when the ledger is unreliable.
[0056] In step S38, the objective function is solved under the constraints that the line loss of each branch is non-negative (i.e., x≥0, line loss cannot be negative) and the sum of the line losses of each branch equals the measured total line loss of the transformer area (i.e., ∑x=P, energy conservation). This constrained optimization problem can be solved using the Alternating Direction Multiplier Method (ADMM) or the interior point method. When the number of branch segments is small (N≤20), the interior point method is preferred because it converges quickly, typically reaching the preset accuracy after 10 to 30 iterations. When the number of branch segments is large (N>20), ADMM is preferred because it decomposes the global problem into subproblems of each branch segment and solves them in parallel, resulting in higher computational efficiency. After the solution is completed, the estimated actual line loss of each branch segment (i.e., the optimal solution for decision variable x) and the estimated impedance of each branch segment (i.e., the optimal solution for decision variable θ) are obtained.
[0057] In one embodiment, step S4, which involves obtaining preset judgment conditions and determining abnormal branch segments based on the preset judgment conditions and the estimated actual line loss of each branch segment, includes: S41. Determine the absolute threshold based on the minimum active power resolution of the intelligent fusion terminal that collects the measurement data at the head end of the distribution transformer and the smart meter that collects the measurement data at the user end; use the difference between the estimated actual line loss value and the corresponding theoretical line loss value of each branch segment as the line loss deviation value, where the theoretical line loss value is the line loss value calculated according to Joule's law based on the time-varying impedance and the corresponding soft estimate of the branch current; determine whether the absolute value of the line loss deviation value is greater than the absolute threshold. If so, the absolute deviation determination result is satisfied; otherwise, the absolute deviation determination result is not satisfied.
[0058] S42. Based on the comprehensive uncertainty of the theoretical line loss value obtained by error propagation calculation of the manufacturing tolerance of the conductor cross-sectional area and the measurement error of the conductor length in the line ledger parameters, determine the relative threshold; determine whether the ratio of the absolute value of the line loss deviation value to the theoretical line loss value is greater than the relative threshold. If so, determine that the relative deviation determination result is satisfied; otherwise, determine that it is not satisfied.
[0059] In this embodiment, the theoretical line loss value is the dependent variable, and its uncertainty stems from the uncertainty of each input independent variable (conductor cross-sectional area, conductor length, etc.), which is transmitted to the dependent variable through a functional relationship (Joule's law). According to the law of error propagation (i.e., the law of uncertainty propagation), when each input quantity is independent and its standard uncertainty is known, the variance of the output quantity is the sum of the squares of the variances of each input quantity multiplied by their corresponding partial derivatives. The theoretical line loss value is calculated according to Joule's law as follows: ; Among them, the branch segment resistance (ρ is resistivity, L is conductor length, and S is conductor cross-sectional area). When calculating the overall uncertainty, first consider the manufacturing tolerance of the conductor cross-sectional area in the line ledger parameters (e.g., ±3%, corresponding to the standard uncertainty, U). s =0.03S / k where k is the coverage factor, usually taken as 2 corresponding to a 95% confidence level) and the measurement error of the conductor length (e.g., ±5%, corresponding to a standard uncertainty U1=0.05L / k), respectively, to obtain U s And U1.
[0060] Then, calculate the partial derivatives of the theoretical line loss value with respect to S and L: ; Find the partial derivative of the theoretical line loss value with respect to L: ; Finally, according to the error propagation formula: ; The comprehensive standard uncertainty of the theoretical line loss value can then be obtained. If the uncertainty of resistivity ρ corrected for temperature also needs to be considered, this term can be added similarly. In this embodiment, the comprehensive relative uncertainty is... The relative threshold is obtained by multiplying the error by a preset coverage factor (e.g., 1.5). This error propagation calculation ensures that the relative threshold can objectively reflect the systematic deviations caused by the inherent tolerances of the ledger parameters and measurement errors, avoiding misjudging normal deviations within the normal tolerance range as abnormalities.
[0061] S43. Based on the standard uncertainty of the actual line loss estimate obtained by error propagation from the measurement error covariance matrix during the solution process of the constraint optimization model, multiply by a preset confidence factor to determine the statistical threshold; determine whether the absolute value of the line loss deviation is greater than the statistical threshold. If so, determine that the statistical significance determination result is satisfied; otherwise, determine that it is not satisfied.
[0062] S44. When the absolute deviation determination result, the relative deviation determination result, and the statistical significance determination result of the same branch segment are all satisfied, the branch segment is determined to be an abnormal branch segment.
[0063] As described in steps S41-S44 above, the sum of the minimum active power resolution of the smart fusion terminal collecting the measurement data at the distribution transformer head end and the minimum active power resolution of the smart meter collecting the measurement data at the user end is taken as the comprehensive resolution of the measurement system. This comprehensive resolution is then multiplied by a preset resolution safety factor to obtain the absolute threshold. Specifically, both the smart fusion terminal and the smart meter have minimum resolution limitations for power measurement. For example, if the active power resolution of a certain model of smart fusion terminal is 5W and the active power resolution of a certain model of smart meter is 1W, then the comprehensive resolution of the measurement system is 6W. Since the actual line loss estimate is obtained by calculating the difference between the power measurement values at the distribution transformer head end and the user end, the quantization errors of the measurements at both ends will propagate. Therefore, the sum of the two resolutions is taken to reflect the worst-case scenario. For example, if the preset value of the resolution safety factor is 3, then the absolute threshold is 18W. When the line loss deviation is less than the absolute threshold, the deviation may only be caused by the superposition of quantization errors of the smart fusion terminal and the smart meter, and is not a real line loss anomaly, thus directly ruling out the possibility of an anomaly in the branch line segment.
[0064] In step S42, based on the manufacturing tolerance of the conductor cross-sectional area and the measurement error of the conductor length in the line ledger parameters, error propagation calculations are performed on each error component according to Joule's law to obtain the comprehensive relative uncertainty of the theoretical line loss value. This comprehensive relative uncertainty is then multiplied by a preset coverage factor to obtain a relative threshold. The theoretical line loss value is calculated using Joule's law based on the time-varying impedance and the corresponding soft estimate of the branch current. Specifically, the calculation of the theoretical line loss value depends on the conductor resistance in the line ledger parameters, which is determined by both the conductor cross-sectional area and the conductor length. These parameters inherently possess uncertainties; for example, the manufacturing tolerance of the conductor cross-sectional area is typically ±3%, and the measurement error of the conductor length is typically ±5%. After error propagation calculations based on the functional relationship of Joule's law for each error component, the comprehensive relative uncertainty of the theoretical line loss value is obtained, with a typical value of approximately 8% to 12%. If the preset value of the coverage factor is, for example, 1.5, then the relative threshold is approximately 12% to 18%. When the ratio of the line loss deviation value to the theoretical line loss value does not exceed the relative threshold, it indicates that the deviation may only be a systematic deviation caused by the manufacturing tolerance of the conductor cross-sectional area and the measurement error of the conductor length, rather than a real line loss anomaly.
[0065] In step S43, the standard uncertainty of the actual line loss estimate obtained from the measurement error covariance matrix through error propagation during the solution process of the constrained optimization model is multiplied by a preset confidence factor to obtain the statistical threshold. Specifically, the actual line loss estimate is obtained by solving the noisy measurement data using the constrained optimization model. Random noise in the measurement data propagates to the actual line loss estimate through the solution process of the constrained optimization model, giving it statistical uncertainty. According to error propagation theory, the standard uncertainty σ of the actual line loss estimate can be obtained by propagating the measurement error covariance matrix through the Jacobian matrix of the constrained optimization model. The preset confidence factor is, for example, 3, corresponding to a 99.7% confidence level, so the statistical threshold is 3σ. When the line loss deviation value does not exceed the statistical threshold, it indicates that the deviation is not statistically significant and may only be a normal fluctuation caused by random measurement noise, rather than a true line loss anomaly.
[0066] In step S44, for each branch segment, the absolute value of the difference between the actual estimated line loss and the corresponding theoretical line loss is calculated as the line loss deviation value. It is then determined whether the line loss deviation value is simultaneously greater than the absolute threshold, whether the ratio of the line loss deviation value to the theoretical line loss value is greater than the relative threshold, and whether the line loss deviation value is greater than the statistical threshold. When all three conditions are met, the branch segment is identified as an abnormal branch segment. This triple-threshold joint determination mechanism ensures that: the absolute threshold eliminates spurious quantitative deviations caused by insufficient resolution of smart fusion terminals and smart meters; the relative threshold eliminates systematic deviations caused by manufacturing tolerances of conductor cross-sectional area and measurement errors of conductor length; and the statistical threshold eliminates statistical fluctuations caused by random measurement noise. This results in highly reliable determination of abnormal branch segments and effectively reduces the false alarm rate.
[0067] In one embodiment, step S5, which generates the fault type based on the multidimensional abnormal physical characteristics, includes: S51. Obtain the target value or value range of each preset fault type in each feature dimension, and generate the feature signature vector corresponding to each preset fault type. The fault physical feature signature library is composed of multiple feature signature vectors.
[0068] S52. Extract the multidimensional abnormal physical features based on the estimated actual line loss of the abnormal branch segment, the measurement data at the head of the distribution transformer, and the measurement data at the user end.
[0069] S53. Based on the standard deviation of each feature dimension under normal conditions, determine the matching weight of each feature dimension according to the relationship that the weight is equal to the inverse square of the standard deviation and then normalized, and calculate the weighted distance between the multidimensional abnormal physical feature and each feature signature vector according to the matching weight, and determine the preset fault type corresponding to the feature signature vector with the smallest weighted distance as the candidate fault type.
[0070] As another preferred embodiment of this invention, the matching between the multidimensional abnormal physical features and the feature signature vectors in the fault physical feature signature library can also be achieved using a pre-trained classification neural network model. This classification neural network model uses the measured values of each feature dimension in the multidimensional abnormal physical features as input vectors, each preset fault type as the output category, and a Softmax function to output the matching probability of the multidimensional abnormal physical features belonging to each preset fault type. The classification neural network model is pre-trained in the following way: A historical line loss anomaly sample library is obtained, containing historical anomaly records of confirmed fault types, each record containing the corresponding multidimensional abnormal physical features and labeled fault type; the network parameters are iteratively updated using gradient descent with the goal of minimizing classification error, until the loss function converges. One or more preset fault types with the highest matching probability output by the classification neural network model are used as candidate fault types, replacing or combining the weighted distance matching results in the output. Preferably, the classification neural network model can employ any one of gradient boosting trees, support vector machines, or an attention-based Transformer architecture. When using a Transformer architecture, the feature dimensions of the multidimensional abnormal physical features are arranged sequentially as input sequences. The Transformer's self-attention mechanism is used to capture the nonlinear coupling relationships between the feature dimensions, thereby improving classification accuracy under conditions of ambiguous feature boundaries. In cases where the output of the classification neural network model differs from the weighted distance matching result, confidence fusion is performed using the output of the classification neural network model as the primary factor and the weighted distance matching result as a secondary factor. The fused result is then used as the candidate fault type.
[0071] S54. Obtain the residual threshold and deviation threshold. Based on the comparison result of Kirchhoff's Current Law residual and residual threshold in the multidimensional abnormal physical characteristics, determine whether to exclude the signature for electricity theft. Based on the comparison result of the neutral current deviation ratio and deviation threshold, determine whether to exclude the signature for leakage. Filter the candidate fault types to obtain the fault type.
[0072] In this embodiment, the Kirchhoff's current law residual sequence of the transformer substation to which the abnormal branch segment belongs during normal operation is collected, and the sample standard deviation of the residual sequence is calculated. This sample standard deviation is used as the standard uncertainty of the Kirchhoff's current law residual, and the standard uncertainty is multiplied by a preset confidence factor to obtain the residual threshold. The neutral current deviation ratio sequence of the transformer substation during normal operation is collected, and the sample standard deviation of the deviation ratio sequence is calculated. This sample standard deviation is used as the standard uncertainty of the neutral current deviation ratio, and the standard uncertainty is multiplied by a preset confidence factor to obtain the deviation threshold. If the transformer substation is a newly commissioned substation without normal operation history data, the comprehensive uncertainty calculated according to the error propagation formula based on the current measurement accuracy of the current transformer at the transformer head end and the smart meter at the user end is used as the residual threshold, and the standard uncertainty determined by the measurement accuracy of the neutral current transformer is used as the deviation threshold.
[0073] As described in steps S51-S54 above, the core idea of fault tracing is that different types of faults have essential differences in physical mechanisms. These differences will leave different "marks" on multiple observable physical quantities. This application formalizes these "marks" into multidimensional abnormal physical features, and formalizes the theoretical "marks" of each fault type derived in advance based on physical laws into a fault physical feature signature library, and realizes the identification of fault types through matching.
[0074] In step S51, a fault physical feature signature library is constructed. The preset fault types cover the main fault modes in low-voltage distribution transformer areas that cause gradual abnormal changes in line losses, including: electricity theft, leakage, poor contact, distributed photovoltaic backfeeding, and metering faults, totaling five categories. It should be noted that other fault modes (such as open circuits and phase-to-phase short circuits) typically cause protection devices to trip rather than gradually abnormally change line losses, and therefore are not included in this signature library. For each preset fault type, based on at least two of Kirchhoff's current law, Ohm's law, and the Joule heating effect, its theoretical target value or range of values in each feature dimension is derived.
[0075] Taking electricity theft faults as an example: According to Kirchhoff's Current Law, electricity theft is equivalent to the existence of an unmetered bypass load on a branch line segment, causing the current flowing into that branch line segment to be greater than the sum of the currents of all metered users. Therefore, the residual of Kirchhoff's Current Law is positive and significantly greater than zero. According to the Joule heating effect, line loss is proportional to the square of the current. The additional current caused by electricity theft will make the line loss deviation strongly positively correlated with the square of the current. According to Ohm's Law, electricity theft does not change the impedance of the line itself, so the impedance deviation approaches zero. Based on the above physical derivations, the characteristic signature vector of electricity theft faults is: Kirchhoff's Current Law residual > 0 (significant), the target value of the correlation coefficient between deviation and the square of current is 1, the target value of the correlation coefficient between deviation and voltage is 0, the target value of impedance deviation is 0, and the target value of the neutral current deviation ratio is within the normal range.
[0076] Taking leakage faults as an example: According to Kirchhoff's Current Law, leakage current flows back through the ground, causing an imbalance between the phase current and the neutral current, with the neutral current deviation ratio significantly deviating from the normal value. According to Ohm's Law, leakage is equivalent to adding an additional resistor in parallel between the phase line and the ground, resulting in a reduction in equivalent impedance and a negative impedance deviation. Based on the above physical derivations, the characteristic signature vector of leakage faults is: neutral current deviation ratio > preset deviation threshold (significant), impedance deviation < 0, Kirchhoff's Current Law residual target value close to zero (because the leakage current still passes through the phase line), and the deviation-voltage correlation coefficient target value is positively correlated (the higher the voltage, the greater the leakage).
[0077] Taking poor contact faults as an example: According to Ohm's law, poor contact leads to additional contact resistance at the joint, increasing the equivalent impedance of the branch segment, resulting in a positive impedance deviation. According to the Joule heating effect, the loss on the additional contact resistance is proportional to the square of the current, and the contact resistance further deteriorates with increasing temperature (positive feedback effect). Therefore, the deviation is positively correlated with ambient temperature, and the time trend is gradual. Based on the above physical derivations, the characteristic signature vector for poor contact faults is: impedance deviation > 0 (significant), target value of the correlation coefficient between deviation and the square of current is 1, target value of the deviation time trend is gradual, and target value of the correlation coefficient between deviation and temperature is positively correlated.
[0078] Taking distributed photovoltaic (PV) backfeeding faults as an example: Distributed PV systems feed power back to the grid during sunshine hours, causing a decrease or even reversal of the net current on the branch lines. The line loss deviation is negative during sunshine hours (actual line loss is less than theoretical line loss), and returns to normal at night when there is no sunshine. Therefore, the deviation exhibits a clear characteristic of "negative deviation during the day and normal at night," and the deviation is negatively correlated with the intensity of sunshine.
[0079] Taking metering faults as an example: metering deviations cause the user's terminal measurement data to be systematically too large or too small, resulting in a systematic deviation in the measured bus loss of the transformer area. However, this deviation is not significantly correlated with physical quantities such as current square, voltage, and temperature, and the residual of Kirchhoff's current law and the neutral current deviation ratio are both within the normal range.
[0080] The feature signature vectors of the above-mentioned preset fault types are summarized to form the fault physical feature signature library.
[0081] In step S52, multidimensional abnormal physical features are extracted. Based on the estimated actual line loss of the abnormal branch segment, the measurement data at the transformer head end, and the measurement data at the user end, at least five of the following multidimensional abnormal physical features are extracted: line loss deviation magnitude (i.e., the absolute value of the difference between the estimated actual line loss and the theoretical line loss), correlation coefficient between deviation and current square (Pearson correlation coefficient calculated for the deviation sequence and the current square sequence within a time window), correlation coefficient between deviation and voltage, correlation coefficient between deviation and ambient temperature data, neutral current deviation ratio (the difference between the measured neutral current and the theoretical calculated value divided by the theoretical calculated value), Kirchhoff's current law residual (the measured current at the branch segment head end minus the sum of currents of all connected users), deviation time trend (the slope obtained by linear regression of the deviation sequence), and deviation time period (the difference between the mean deviation during the day and the mean deviation during the night).
[0082] The reason for selecting at least five feature dimensions is that: this solution pre-determines five types of faults. According to pattern recognition theory, the number of feature dimensions must be no less than the number of categories to achieve effective differentiation. At the same time, selecting at least five of the above eight features can ensure sufficient differentiation capability even when some feature data is missing (such as the neutral current deviation ratio being unavailable because a neutral current sensor is not installed in a certain area).
[0083] In step S53, weighted matching is performed. Since the dimensions and fluctuation ranges of each feature dimension are different, normalized weighting is required. Based on the standard deviation σ of each feature dimension under normal conditions, the matching weight w of each feature dimension is determined according to the relationship that the weight equals the inverse square of the standard deviation before normalization, i.e.: ; Among them, w m σ represents the matching weight of the m-th feature dimension; σ represents the standard deviation of the m-th feature dimension under normal conditions; M represents the total number of feature dimensions. The physical meaning of this weight allocation method is: the feature dimension with smaller fluctuations under normal conditions has stronger discriminative ability and is assigned a higher weight.
[0084] Regarding the method of obtaining σ: σ is estimated from the standard deviation of the most recent 30 days of normal operation data (excluding confirmed abnormal periods). The determination of the normal operation period is automatically determined through the following data filtering and steady-state discrimination process, without the need for manual labeling: For the collected distribution transformer head-end measurement data and user end measurement data, the following cleaning operation is first performed: First, an integrity check is performed: the data acquisition success rate for each sampling period is checked. If the data loss rate of any measurement channel exceeds 5% within a certain time period, that time period is marked as "incomplete data" and excluded. For sporadic missing data points, linear interpolation of adjacent sampling points is used to fill in the gaps. If more than 3 sampling points are missing consecutively in the same measurement channel (i.e., no valid data for 45 consecutive minutes), that time period is excluded entirely and not included in the calculation of the normal operating baseline.
[0085] Then, a rationality check is performed: each measured data point is checked to ensure it is within the physically reasonable range. Data is considered invalid if the voltage value exceeds ±20% of the rated voltage (i.e., below 176V or above 264V); if the current value is negative; if the active power exceeds ±50% of the user's applied capacity; or if the power factor is less than -1 or greater than 1. If any measurement channel contains any of these invalid data points, the entire sampling point will not be used.
[0086] Finally, abrupt change point detection: The voltage, current, and power sequences are differentially detected point by point. If the power change rate between two adjacent sampling points exceeds 30% of the user's installed capacity, that point is determined to be a load abrupt change point. Since the switching of high-power equipment or sudden changes in distributed photovoltaic output can cause instantaneous fluctuations in line loss rate, such abrupt change points and one sampling point before and after them are excluded to avoid transient processes contaminating the statistical benchmark for normal operation.
[0087] After the above cleaning, using 15-minute measurement data as a sampling point, the overall operating status of the transformer area is distinguished between steady-state and transient states: a continuous 24-hour period (i.e., 96 sampling points) is taken as a discrimination window, and the difference between the maximum and minimum line loss rate of the transformer area within this window (i.e., the range) is calculated as a measure of the fluctuation amplitude of the line loss rate within the window. If the range is less than 10% of the rated line loss rate of the transformer area (for example, if the rated line loss rate of a transformer area is 6%, then the threshold is 0.6 percentage points), then the transformer area is determined to be in steady-state operation within the window; if the range is greater than or equal to the threshold, then a transient process exists within the window (such as the concentrated start-up and shutdown of multiple high-power devices, or drastic fluctuations in the output of distributed photovoltaic systems), and the corresponding time period of the window is excluded. From the remaining time periods after excluding transient windows, the segment with the longest continuous steady-state operation time, and whose duration is not less than 72 hours (i.e., 288 continuous sampling points), is selected as the candidate normal operation period.
[0088] For candidate normal operating periods, the following three confirmation checks are performed to ensure their "normal" attribute is reliable: Overall line loss rate check: Calculate the average line loss rate of the distribution area within this period. If the average is less than or equal to zero or greater than or equal to 15%, the period is excluded. A line loss rate less than or equal to zero indicates potential power backflow due to large-scale distributed photovoltaic power generation or metering device errors. A line loss rate greater than or equal to 15% clearly exceeds the normal line loss range of 2% to 10% for low-voltage distribution areas, and neither qualifies as a "normal" benchmark. Electrical balance check: For each sampling point within the candidate period, check whether Kirchhoff's current law is satisfied—that is, whether the current of each phase at the transformer's head end is approximately equal to the sum of the currents of all users carried by that phase. If the current imbalance deviation at a sampling point exceeds 2% of the rated current at the transformer's head end, that sampling point is removed; if the remaining sampling points after removal are less than 90% of the original total, the entire candidate period is excluded. Continuous duration verification: After the above two checks, if the remaining continuous normal period is less than 72 hours, the range threshold in the steady-state discrimination should be appropriately relaxed (relaxed to 1.5 times the original threshold), and the steady-state discrimination and secondary confirmation should be performed again; if more than 72 hours of continuous normal period cannot be obtained after relaxing it three times, the area should be marked as "insufficient reference data", and historical data of the same area and the same type (same conductor model, similar number of users) should be used as the substitute reference.
[0089] After the above three-layer screening and confirmation, all time periods that passed all inspections within the most recent 30 calendar days are taken as the "normal operation data" for calculating the standard deviation of each characteristic dimension. Throughout the screening process, all thresholds (data missing rate 5%, voltage deviation ±20%, power change rate 30%, line loss rate range threshold 10%, KCL deviation 2%, continuous duration 72 hours, etc.) can be configured and adjusted by maintenance personnel in the system according to the actual operating characteristics of the transformer area to adapt to the power grid characteristics of different areas. For newly commissioned transformer areas without historical data, historical data from transformer areas of the same type (same conductor type, similar number of users, same power supply area) can be used as a substitute; if data from transformer areas of the same type is also unavailable, then 5% of the range of each characteristic dimension is taken as the default standard deviation.
[0090] The weighted distance d between the multidimensional anomaly physical feature f and each feature signature vector s is calculated based on the matching weight: ; Where, d k f represents the weighted distance between the multidimensional abnormal physical features and the feature signature vector of the k-th preset fault type; m s represents the measured value of the m-th feature dimension in the multidimensional anomaly physical characteristics; k,m w represents the target value of the m-th feature dimension in the feature signature vector of the k-th preset fault type; mThis represents the matching weight for the m-th feature dimension. The preset fault type corresponding to the feature signature vector with the smallest weighted distance is determined as the candidate fault type.
[0091] In step S54, hard rule filtering is performed on the candidate fault types. Hard rule filtering uses deterministic criteria based on physical laws to exclude physically impossible diagnostic results. Specifically, it includes: if the residual of Kirchhoff's current law in the multidimensional abnormal physical characteristics is less than the residual threshold determined based on the standard deviation of measurement error (i.e., Kirchhoff's current law is basically balanced), then the signature for electricity theft is excluded, because electricity theft inevitably leads to an imbalance in Kirchhoff's current law; if the neutral current deviation ratio in the multidimensional abnormal physical characteristics is less than the deviation threshold determined based on the fluctuation of the neutral current during normal operation (i.e., the neutral current is normal), then the signature for leakage current is excluded, because leakage current inevitably leads to an abnormal neutral current. After hard rule filtering, the remaining candidate fault types are taken as the final fault types.
[0092] Simultaneously, an anomaly evidence chain is generated based on the matching results. The anomaly evidence chain records the measured value of each feature component, the corresponding physical law basis, and the direction of contribution of that feature component to the final diagnostic conclusion.
[0093] In one embodiment, step S6, which involves obtaining the spatial location of the fault on the abnormal branch segment based on the estimated actual line loss of the abnormal branch segment, includes: S61. Obtain the impedance estimate of the abnormal branch segment based on the solution results of the constraint optimization model; S62. Calculate the theoretical impedance value at the operating temperature based on the ambient temperature data and the estimated branch current value of the abnormal branch segment, and subtract the theoretical impedance value from the estimated impedance value to obtain the impedance deviation after deducting the temperature effect. S63. Calculate the equivalent distance from the fault point to the beginning of the branch based on the impedance deviation and the resistance per unit length of the abnormal branch segment. S64. Obtain the deviation time trend based on the multidimensional abnormal physical characteristics, and correct the equivalent distance based on the deviation time trend: when the time trend is a gradual trend, find the nearest joint position record in the line ledger parameters based on the equivalent distance, and take the position range corresponding to the joint position record as the spatial position; when the time trend is a sudden trend, directly take the equivalent distance as the spatial position.
[0094] As another preferred embodiment of this invention, to further improve the estimation accuracy of the spatial location, a deviation correction model trained based on historical fault data can be introduced. Specifically, on-site repair records of confirmed faults are collected. Each record contains the positioning deviation value between the equivalent distance and the actual fault location, as well as the corresponding deviation time trend and equivalent distance. Using the equivalent distance and deviation time trend as input features, and the deviation value between the actual fault location and the equivalent distance as labels, the deviation correction model is trained. The deviation correction model uses a random forest regression model or a lightweight gradient booster (LightGBM) and is trained with mean squared error as the loss function. After training, the equivalent distance and deviation time trend of the current abnormal branch segment are input into the deviation correction model, and the correction coefficient is output. The correction coefficient is multiplied by the equivalent distance to obtain the corrected spatial location estimate. This correction scheme is particularly suitable for scenarios where the time trend is a gradual trend and the joint location record in the line ledger parameters is incomplete or lacks accuracy, compensating for the impact of ledger parameter errors on positioning accuracy through data-driven correction.
[0095] As described in steps S61-S64 above, the core idea of fault location is that a fault (such as poor contact, broken strand, etc.) will introduce additional resistance on the branch line segment. The magnitude of the additional resistance is proportional to the distance of the fault point from the beginning. Therefore, the spatial location of the fault can be deduced by estimating the magnitude of the additional resistance.
[0096] In step S61, the impedance estimate of the abnormal branch segment is obtained based on the solution results of the constrained optimization model. In the constrained optimization model, the impedance parameter vector θ participates in the optimization solution as a decision variable. After the solution is completed, the corresponding component of the abnormal branch segment is extracted from the optimal solution of the impedance parameter vector, which is the impedance estimate of that branch segment. This impedance estimate reflects the actual equivalent resistance of the branch segment under the current operating conditions.
[0097] In step S62, the impedance deviation after deducting the temperature effect is calculated. Since the conductor resistance itself changes with temperature, the impedance estimate includes the resistance change caused by normal temperature effects, which needs to be deducted to separate the additional resistance caused by the fault. Specifically, based on the ambient temperature data and the estimated branch current of the abnormal branch segment, and according to the current conductor temperature after the coupling iteration convergence in step S2, the theoretical impedance value at the operating temperature is calculated, i.e.: ; Among them, R theory R represents the theoretical impedance value at the operating temperature, in Ω. 20 The basic impedance value is calculated from the line ledger parameters and is expressed in Ω; α represents the temperature coefficient of resistance; T condThis represents the current conductor temperature after the coupling iteration converges, determined by the iterative process in step S2, and is expressed in °C.
[0098] Subtracting the theoretical impedance value from the impedance estimate yields the impedance deviation after accounting for the temperature effect: ; Where, ΔR fault This represents the impedance deviation after deducting the temperature effect, in Ω. The impedance estimate of the abnormal branch segment is obtained from the solution of the constrained optimization model, and the unit is Ω; R theory This represents the theoretical impedance value at operating temperature, calculated using the previous formula, and is expressed in Ω. A positive impedance deviation indicates the presence of additional resistance (e.g., poor contact, broken strand); a negative impedance deviation indicates the presence of a parallel bypass (e.g., leakage).
[0099] In step S63, the equivalent distance from the fault point to the beginning of the branch is calculated based on the impedance deviation and the resistance per unit length of the abnormal branch segment: ; Among them, l fault ΔR represents the equivalent distance from the fault point to the beginning of the branch, in meters. fault The impedance deviation after deducting the temperature effect is calculated using the previous formula, with units of Ω; r0 represents the resistance per unit length of the abnormal branch segment, calculated from the conductor resistivity and cross-sectional area in the line ledger parameters (r=ρ / S), with units of Ω / m. For example, if the impedance deviation of an abnormal branch segment is 0.012Ω and the resistance per unit length is 0.0004Ω / m, then the equivalent distance is 0.012 / 0.0004=30m, meaning the fault point is approximately 30 meters from the beginning of the branch segment.
[0100] In step S64, the equivalent distance is corrected based on the deviation time trend. The deviation time trend originates from the deviation time trend component in the multidimensional abnormal physical characteristics. When the time trend is a gradual trend (i.e., the deviation gradually increases over several days or weeks), it indicates that the fault is a progressive process (such as joint oxidation, bolt loosening, etc.), and such faults usually occur at conductor joints. Therefore, the nearest joint location record is found in the line ledger parameters based on the equivalent distance (the line ledger parameters usually record the location of each joint), and the location range corresponding to the joint location record is used as the spatial location. For example, if the equivalent distance is 30 meters, and the line ledger parameters record that there is a joint at 28 meters and 45 meters for this branch segment, then the joint at 28 meters is selected as the spatial location. When the time trend is a sudden trend (i.e., the deviation suddenly appears in a short period of time), it indicates that the fault is a sudden event (such as external damage, lightning strikes breaking strands, etc.), and such faults may occur at any location on the branch segment. Therefore, the equivalent distance is directly used as the spatial location.
[0101] In one embodiment, step S7, where the output includes the source location results of the abnormal branch segment, the fault type, the spatial location, and the abnormal evidence chain, specifically includes: The source tracing and location results are output in the form of a structured report, which includes the following fields: abnormal branch segment number and name, fault type, spatial location (distance from the branch head or joint number), abnormal evidence chain (measured values of each characteristic component, physical law basis, and contribution direction), actual line loss estimate, theoretical line loss value, deviation range, and confidence level. Simultaneously, the source tracing and location results are pushed to the power supply company's operation and maintenance management system to generate a corresponding inspection work order. The inspection work order carries the geographical coordinates and spatial location information of the abnormal branch segment, enabling maintenance personnel to directly navigate to the fault point for on-site troubleshooting.
[0102] like Figure 2 As shown, the present invention also provides a low-voltage branch line loss anomaly tracing and location system, comprising: The partitioning module is used to acquire the topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data and ambient temperature data of the low-voltage distribution transformer area, and construct a radial tree network model based on the topology file to divide the low-voltage distribution transformer area into multiple branch segments. The allocation module is used to obtain the time-varying impedance of each branch line segment based on the line ledger parameters and ambient temperature data, and to construct the current distribution relationship of each branch line segment according to Kirchhoff's current law. The actual line loss estimation module is used to establish a constrained optimization model based on the measurement data at the head end of the distribution transformer and the measurement data at the end of each user; and to solve the constrained optimization model to obtain the actual line loss estimation value of each branch segment. An abnormal branch determination module is used to obtain preset determination conditions and determine abnormal branch segments based on the preset determination conditions and the estimated actual line loss of each branch segment. The fault type generation module is used to extract multi-dimensional abnormal physical features based on the actual line loss estimate of the abnormal branch segment, the measurement data at the head end of the distribution transformer and the measurement data at the user end, and to generate fault types and abnormal evidence chains based on the multi-dimensional abnormal physical features. The spatial location acquisition module is used to obtain the spatial location of the fault on the abnormal branch segment based on the estimated actual line loss of the abnormal branch segment; The result generation module is used to generate source tracing and location results based on the abnormal branch segments, fault types, spatial locations, and abnormal evidence chains.
[0103] In one embodiment, the allocation module includes: The basic impedance value acquisition unit is used to acquire the basic impedance value of each branch line segment at the reference temperature based on the line ledger parameters. The current conductor temperature acquisition unit is used to calculate the current conductor temperature based on the ambient temperature data and the estimated branch current obtained from the previous iteration, according to the linear superposition relationship between conductor temperature and the ambient temperature and the square of the branch current. The branch impedance value acquisition unit is used to correct the basic impedance value based on the current conductor temperature and the temperature coefficient of resistance to obtain the branch impedance value under the current iteration number. The branch current estimation unit is used to re-execute the solution of the constraint optimization model based on the branch impedance value to obtain the updated branch current estimation value. The judgment unit is used to determine whether the convergence criterion is met based on the updated branch current estimate, wherein the convergence criterion is that the absolute value of the change in the current conductor temperature between two adjacent iterations is less than a first preset threshold and the relative change in the branch impedance value between two adjacent iterations is less than a second preset threshold. If the convergence criterion is met, the branch impedance value at the current iteration number is output as the time-varying impedance; if the convergence criterion is not met and the iteration number does not exceed the preset maximum iteration number, the branch current estimate obtained based on the ambient temperature data and the previous iteration is returned.
[0104] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent results or equivalent process transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. A method for tracing and locating abnormal low-voltage branch line losses, applied to low-voltage distribution transformer areas, characterized in that, include: The topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data and ambient temperature data of the low-voltage distribution transformer area are obtained, and a radial tree network model is constructed based on the topology file to divide the low-voltage distribution transformer area into multiple branch segments. The time-varying impedance of each branch line segment is obtained based on the line ledger parameters and ambient temperature data, and the current distribution relationship of each branch line segment is constructed according to Kirchhoff's current law. A constrained optimization model is established based on the measurement data at the head end of the distribution transformer and the measurement data at each user end; the constrained optimization model is solved to obtain the estimated actual line loss of each branch segment; Obtain preset judgment conditions, and determine abnormal branch segments based on the preset judgment conditions and the estimated actual line loss of each branch segment; Based on the estimated actual line loss of the abnormal branch segment, the measurement data at the head end of the distribution transformer and the measurement data at the user end, multi-dimensional abnormal physical features are extracted, and fault types and abnormal evidence chains are generated based on the multi-dimensional abnormal physical features. The spatial location of the fault on the abnormal branch segment is obtained based on the estimated actual line loss of the abnormal branch segment; The source tracing and location results are generated based on the abnormal branch segments, fault types, spatial locations, and abnormal evidence chains.
2. The method for tracing and locating abnormal low-voltage branch line losses according to claim 1, characterized in that, The step of obtaining the time-varying impedance of each branch segment based on the line ledger parameters and ambient temperature data includes: The basic impedance value of each branch segment at the reference temperature is obtained based on the line ledger parameters. Based on the ambient temperature data and the branch current estimate obtained from the previous iteration, the current conductor temperature is calculated according to the linear superposition relationship between conductor temperature and the ambient temperature and the square of the branch current. The base impedance value is corrected based on the current conductor temperature and temperature coefficient of resistance to obtain the branch impedance value at the current iteration number; Based on the branch impedance value, the constraint optimization model is solved again to obtain the updated branch current estimate. The convergence criterion is determined based on the updated branch current estimate, wherein the convergence criterion is that the absolute value of the current conductor temperature change in two adjacent iterations is less than a first preset threshold and the relative change of the branch impedance value in two adjacent iterations is less than a second preset threshold. If the convergence criterion is met, the branch impedance value at the current iteration number is output as the time-varying impedance; if the convergence criterion is not met and the iteration number does not exceed the preset maximum iteration number, the process returns to executing the branch current estimate obtained from the ambient temperature data and the previous iteration.
3. The method for tracing and locating abnormal low-voltage branch line losses according to claim 1, characterized in that, The step of establishing a constrained optimization model based on the measurement data at the head end of the distribution transformer and the measurement data at each user end includes: Based on the active power, reactive power and voltage measured at each user terminal, the current estimate of each user is calculated according to the inverse relationship that current equals the ratio of apparent power to voltage. Then, based on the radial tree network model, the current estimate of each user is summarized according to the branch segments to obtain the soft estimate of the branch current of each branch segment. The measured bus loss of the distribution area is obtained by subtracting the total power of all users from the total power of the distribution transformer head end measurement data in the distribution transformer head end measurement data; Based on the topology of the radial tree network model, an equality constraint is generated to ensure that the sum of the branch currents of each branch segment equals the total current at the head of the transformer. The soft estimate of the branch current and the measured bus loss of the transformer area are used as the measurement inputs of the constraint optimization model, and the equality constraint is used as the structural constraint of the constraint optimization model.
4. The method for tracing and locating abnormal low-voltage branch line losses according to claim 3, characterized in that, The step of obtaining preset judgment conditions and determining abnormal branch segments based on the preset judgment conditions and the estimated actual line loss of each branch segment includes: Based on the minimum active power resolution of the intelligent fusion terminal that collects the measurement data at the head end of the distribution transformer and the smart meter that collects the measurement data at the user end, an absolute threshold is determined; the difference between the actual line loss estimate and the corresponding theoretical line loss value of each branch segment is taken as the line loss deviation value, where the theoretical line loss value is the line loss value calculated according to Joule's law based on the time-varying impedance and the corresponding soft estimate of the branch current; it is then determined whether the absolute value of the line loss deviation value is greater than the absolute threshold. If it is, the absolute deviation determination result is satisfied; otherwise, the absolute deviation determination result is not satisfied. Based on the comprehensive uncertainty of the theoretical line loss value obtained by error propagation calculation of the manufacturing tolerance of conductor cross-sectional area and the measurement error of conductor length in the line ledger parameters, a relative threshold is determined; it is then determined whether the ratio of the absolute value of the line loss deviation value to the theoretical line loss value is greater than the relative threshold. If so, the relative deviation determination result is determined to be satisfied; otherwise, it is determined to be unsatisfactory. The standard uncertainty of the actual line loss estimate obtained from the constrained optimization model is multiplied by a preset confidence factor to determine the statistical threshold; it is then determined whether the absolute value of the line loss deviation is greater than the statistical threshold. If so, the statistical significance determination result is deemed to be satisfied; otherwise, it is deemed not to be satisfied. When the absolute deviation determination result, the relative deviation determination result, and the statistical significance determination result of the same branch segment are all satisfied, the branch segment is determined to be an abnormal branch segment.
5. The method for tracing and locating abnormal low-voltage branch line losses according to claim 3, characterized in that, The step of generating fault types based on the multidimensional abnormal physical characteristics includes: Obtain the target value or value range of each preset fault type in each feature dimension, and generate the feature signature vector corresponding to each preset fault type. The fault physical feature signature library is composed of multiple feature signature vectors. Based on the estimated actual line loss of the abnormal branch segment, the measurement data at the head of the distribution transformer, and the measurement data at the user end, the multidimensional abnormal physical features are extracted. Based on the standard deviation of each feature dimension under normal conditions, the matching weight of each feature dimension is determined according to the relationship that the weight is equal to the inverse square of the standard deviation and then normalized. Based on the matching weight, the weighted distance between the multidimensional abnormal physical feature and each feature signature vector is calculated, and the preset fault type corresponding to the feature signature vector with the smallest weighted distance is determined as the candidate fault type. Obtain the residual threshold and deviation threshold. Based on the comparison result of Kirchhoff's Current Law residual and residual threshold in the multidimensional abnormal physical characteristics, determine whether to exclude electricity theft signatures. Based on the comparison result of neutral current deviation ratio and deviation threshold, determine whether to exclude leakage signatures. Filter the candidate fault types to obtain the fault types.
6. The method for tracing and locating abnormal low-voltage branch line losses according to claim 5, characterized in that, The step of obtaining the spatial location of the fault on the abnormal branch segment based on the estimated actual line loss of the abnormal branch segment includes: The impedance estimate of the abnormal branch segment is obtained based on the solution results of the constrained optimization model. The theoretical impedance value at the operating temperature is calculated based on the ambient temperature data and the estimated branch current of the abnormal branch segment. The theoretical impedance value is then subtracted from the estimated impedance value to obtain the impedance deviation after deducting the temperature effect. Based on the impedance deviation and the resistance per unit length of the abnormal branch segment, calculate the equivalent distance from the fault point to the beginning of the branch. The deviation time trend is obtained based on the multidimensional abnormal physical characteristics, and the equivalent distance is corrected based on the deviation time trend: when the time trend is a gradual trend, the nearest joint position record is found in the line ledger parameters based on the equivalent distance, and the position range corresponding to the joint position record is used as the spatial position; when the time trend is a sudden trend, the equivalent distance is directly used as the spatial position.
7. A low-voltage branch line loss anomaly tracing and location system, characterized in that, include: The partitioning module is used to acquire the topology file, line ledger parameters, transformer head-end measurement data, user terminal measurement data and ambient temperature data of the low-voltage distribution transformer area, and construct a radial tree network model based on the topology file to divide the low-voltage distribution transformer area into multiple branch segments. The allocation module is used to obtain the time-varying impedance of each branch line segment based on the line ledger parameters and ambient temperature data, and to construct the current distribution relationship of each branch line segment according to Kirchhoff's current law. The actual line loss estimation module is used to establish a constrained optimization model based on the measurement data at the head end of the distribution transformer and the measurement data at the end of each user; and to solve the constrained optimization model to obtain the actual line loss estimation value of each branch segment. An abnormal branch determination module is used to obtain preset determination conditions and determine abnormal branch segments based on the preset determination conditions and the estimated actual line loss of each branch segment. The fault type generation module is used to extract multi-dimensional abnormal physical features based on the actual line loss estimate of the abnormal branch segment, the measurement data at the head end of the distribution transformer and the measurement data at the user end, and to generate fault types and abnormal evidence chains based on the multi-dimensional abnormal physical features. The spatial location acquisition module is used to obtain the spatial location of the fault on the abnormal branch segment based on the estimated actual line loss of the abnormal branch segment; The result generation module is used to generate source tracing and location results based on the abnormal branch segments, fault types, spatial locations, and abnormal evidence chains.
8. A low-voltage branch line loss anomaly tracing and location system according to claim 7, characterized in that, The allocation module includes: The basic impedance value acquisition unit is used to acquire the basic impedance value of each branch line segment at the reference temperature based on the line ledger parameters. The current conductor temperature acquisition unit is used to calculate the current conductor temperature based on the ambient temperature data and the estimated branch current obtained from the previous iteration, according to the linear superposition relationship between conductor temperature and the ambient temperature and the square of the branch current. The branch impedance value acquisition unit is used to correct the basic impedance value based on the current conductor temperature and the temperature coefficient of resistance to obtain the branch impedance value under the current iteration number. The branch current estimation unit is used to re-execute the solution of the constraint optimization model based on the branch impedance value to obtain the updated branch current estimation value. The judgment unit is used to determine whether the convergence criterion is met based on the updated branch current estimate, wherein the convergence criterion is that the absolute value of the change in the current conductor temperature between two adjacent iterations is less than a first preset threshold and the relative change in the branch impedance value between two adjacent iterations is less than a second preset threshold. If the convergence criterion is met, the branch impedance value at the current iteration number is output as the time-varying impedance; if the convergence criterion is not met and the iteration number does not exceed the preset maximum iteration number, the branch current estimate obtained based on the ambient temperature data and the previous iteration is returned.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.
