Physical constraint guided power distribution network multi-source data fusion and state completion method and device

Abstract

This invention discloses a method and apparatus for physical constraint-guided multi-source data fusion and state completion in distribution networks, comprising: acquiring and preprocessing multi-source heterogeneous operating data; constructing an observation tensor, an observation mask tensor, and a weight tensor; establishing a state completion model containing low-rank constraints, weighted observation consistency constraints, and physical consistency constraints constructed from linearized power flow and power balance; iteratively solving the model using the alternating direction multiplier method to obtain the completed state tensor; and generating and outputting a confidence tensor and bad data markers based on observation residuals and physical residuals. This invention can output physically consistent network-wide state data under conditions of low observability and multi-source heterogeneity, reducing dependence on measurement redundancy and communication bandwidth, and improving the availability and reliability of distribution network state data.

Description

Technical Field

[0001] This invention relates to the field of data processing and operational status sensing technology for power system distribution networks, specifically to a method for multi-source data fusion and status completion for distribution networks under low observability conditions, and also to a device for implementing this method. Background Technology

[0002] With the increasing scale of distributed power sources, energy storage, and controllable loads connected to distribution networks, the operating status of these networks exhibits greater time-varying and uncertainties. State assessment, anomaly analysis, and operational optimization of distribution networks typically require obtaining key state variables such as voltage and power across the entire network. However, in practical engineering, the deployment of measurement devices is often limited by cost, geographical conditions, and maintenance requirements, resulting in low measurement coverage and long-term unmeasurability of some nodes or branches. Simultaneously, communication links may experience packet loss and latency fluctuations, leading to missing data segments and inconsistent arrival times. Furthermore, differences in accuracy, sampling frequency, and timestamp synchronization quality among multi-source data make it difficult to directly align and merge state variables at the same moment. These factors collectively result in distribution networks being in a state of low or weak observability, making it difficult to generate continuous, complete, and reliable network-wide state data.

[0003] In existing technologies, classical state estimation methods often rely on high measurement redundancy and stable data quality. When measurement coverage is insufficient or the proportion of missing data is high, the estimation model may encounter problems such as unobservability, difficulty in convergence, or sensitivity to bad data, making it difficult to stably output the overall network state. For data completion, commonly used linear interpolation, spline interpolation, or historical mean-based methods are simple to implement, but they typically only utilize local time-series information, making it difficult to leverage inter-node correlations and not guaranteeing that the completed data satisfies the power flow and balance laws of the distribution network. Some technologies employ statistical regression or low-rank decomposition to utilize spatiotemporal correlations, but without constraints on electrical and physical laws, the completed data may be inconsistent with line parameters and power balance, making it difficult to directly use the completed data for subsequent state assessment and operational analysis.

[0004] Therefore, there is an urgent need for a method that can achieve unified fusion of multi-source heterogeneous data under low observability conditions, and introduce physical consistency constraints such as power flow and power balance of distribution networks during the completion process, so as to output interpretable, physically consistent, and directly applicable network status data. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a method and apparatus for physical constraint-guided fusion and state completion of multi-source sparse data in distribution networks. This method is used to recover high spatiotemporal resolution state data of the entire network under conditions of low measurement coverage, missing data, and significant quality differences, and provides confidence level and bad data labeling.

[0006] A method for physical constraint-guided multi-source data fusion and state completion in distribution networks includes the following steps:

[0007] S1: Acquire multi-source heterogeneous operation data of the distribution network and perform preprocessing. The multi-source heterogeneous operation data includes at least electrical measurement data of some nodes or branches and external auxiliary data related to the electrical measurement data. The preprocessing includes at least data cleaning, time alignment and feature standardization.

[0008] S2: Constructing the observation tensor based on preprocessed multi-source heterogeneous operational data Observation mask tensor with weight tensor The observation tensor The observation mask tensor is used to characterize the measurements at each observation location. The weight tensor is used to mark valid observation locations. Used to characterize the reliability of each observation, and the , , As input to the state completion optimization model in step S3;

[0009] S3: Based on the observed tensor Observation mask tensor with weight tensor Establish the tensor of the state to be completed The state completion optimization model includes at least: a low-rank constraint term for characterizing the correlation of the state tensor in the node dimension and the time dimension, a weighted observation consistency term for constraining the observation location to remain consistent with the original observation, and a physical consistency constraint term for constraining the completion result to satisfy the power flow relationship and power balance relationship of the distribution network linearization. S4: The state completion optimization model is iteratively solved using the alternating direction multiplier method to obtain the completed state tensor. ; S5: Generate a confidence tensor based on observation consistency residuals and physical consistency residuals. Marking suspicious bad data and output the , and .

[0010] Furthermore, the electrical measurement data includes at least one of the following: node voltage amplitude. Node-injected active power Node-injected reactive power Branch active power Branch reactive power Branch current The external auxiliary data includes at least one of the following: meteorological data, daily data, and equipment parameter data; the feature dimensions It consists of electrical measurement characteristics and external auxiliary characteristics.

[0011] Furthermore, the preprocessing in step S1 includes: identifying outliers using box plots, Hample filtering, or Kalman filtering; and processing data from different sampling frequencies at a uniform time step. Resampling is used to achieve time alignment; linear interpolation or forward padding is used for alignment of short-time missing data, while missing location markers are preserved for constructing the observation mask tensor. .

[0012] Furthermore, the weight tensor The observation mask tensor is determined based on at least one of the following: data source type, measurement accuracy level, timestamp synchronization error, link latency, packet loss rate, or historical residual statistics, and then normalized. Satisfy: When In the index When there are observations at a location ,when In the index When missing .

[0013] Furthermore, the objective function of the state completion optimization model described in step S3 is:

[0014]

[0015] in, For low-rank constraint terms, For the weighted observation consistency term, This is a physical consistency constraint. For the tensor nuclear norm, It is the Frobenius norm. Indicates element-wise product. , , These are non-negative weighting coefficients.

[0016] Furthermore, the tensor kernel norm For the state tensor The sum of the nuclear norms after multimodal expansion, or the tensor nuclear norms based on t-SVD, can be used to characterize the correlation of the distribution network state in the node dimension and time dimension and to achieve low-rank constraints.

[0017] Furthermore, the physical consistency constraint term Constructed based on linearized branch power flow model and power balance relationship; for any time t and any branch... ,satisfy , , ,in and Branch roads The resistance and reactance, For nodes The set of downstream child nodes; and constructed using the sum of squares of the residuals of the equation. .

[0018] Furthermore, the alternating direction multiplier method in step S4 includes: splitting the low-rank constraint term and the physical consistency constraint term into different subproblems for alternating updates by introducing auxiliary variables; in each iteration, performing tensor singular value thresholding updates on the low-rank subproblems, performing weighted least squares updates on the subproblems containing weighted observation consistency terms and physical consistency constraints, and stopping the iteration when the original residual and the dual residual satisfy the preset convergence condition.

[0019] Furthermore, the confidence tensor mentioned in step S5 The weighted observation consistency residual, physical consistency residual, and the weight tensor Jointly determined; when the residual corresponding to an observation point meets the preset criteria, the observation point is marked as suspicious bad data and written into the suspicious bad data marker. .

[0020] A state completion device for implementing the method described above includes: a data acquisition and preprocessing module, a tensor construction module, a model construction module, a completion solution module, and a result output module; wherein, the data acquisition and preprocessing module is used to acquire and preprocess multi-source heterogeneous operation data of a power distribution network, the multi-source heterogeneous operation data including at least some electrical measurement data of nodes or branches and external auxiliary data related to the electrical measurement data, and the preprocessing including at least data cleaning, time alignment, and feature standardization;

[0021] The tensor construction module is used to construct observation tensors based on preprocessed multi-source heterogeneous operational data. Observation mask tensor with weight tensor ,in For the number of nodes, For time steps, For feature dimension;

[0022] The model building module is used to establish the state tensor to be completed. The state completion optimization model includes at least a low-rank constraint term, a weighted observation consistency term, and a physical consistency constraint term.

[0023] The completion and solution module is used to iteratively solve the state completion optimization model using the alternating direction multiplier method to obtain the completed state tensor. Based on the observation consistency residuals and physical consistency residuals, a confidence tensor is generated. Marking suspicious bad data and output the , and The technical solution of the present invention has the following advantages compared with the prior art:

[0024] 1. This invention unifies multi-source heterogeneous measurements and external auxiliary information into a third-order tensor. And explicitly construct the mask This allows missing points and missing segments to enter the completion model with a consistent data structure, thus enabling the formation of continuous network-wide status data even under conditions of insufficient measurement coverage and frequent packet loss and delay. Compared to methods based solely on time series interpolation, this invention also utilizes the correlation between nodes and time, making the completion results more adaptable to missing structures.

[0025] This invention introduces a weight tensor Weighted constraints are applied to observations from different sources, with different alignment qualities and different reliability. Observations with higher reliability contribute more to the completion results, while observations with lower reliability are automatically downweighted, thereby reducing the bias risk caused by heterogeneous data fusion. This mechanism enables the completion model to remain stable when the quality of multi-source data fluctuates and reduces the transmission impact of low-quality observations on the recovery of the overall network state.

[0026] 2. This invention introduces physical consistency constraints into the objective function. The completion result is physically corrected by linearizing the power flow and power balance residuals, so that... It conforms to the electrical laws of the distribution network; compared with the pure statistical low-rank completion method, the present invention can significantly reduce the inconsistency between voltage and power completion results and line parameters and power balance, thereby improving the availability of completion data in state estimation, anomaly analysis and operation evaluation.

[0027] 3. This invention uses the alternating direction multiplier method to achieve the split solution of low-rank approximation and physical correction, decomposing the complex multi-objective problem into an efficient subproblem and iteratively converging it; the physical correction subproblem can be written in sparse weighted least squares form, which is convenient to solve using sparse linear algebra, thus ensuring numerical stability while having good scalability, and is suitable for feeders of different sizes and time windows.

[0028] 4. The present invention outputs a confidence tensor. And give a bad data marker. This ensures that the completion results not only include numerical values ​​but also interpretable reliability information; by using a joint criterion of observational residuals and physical residuals, hidden bad data can be identified and closed-loop governance of secondary completion can be supported, thereby improving the maintainability and long-term operational quality of the data base. Attached Figure Description

[0029] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0030] Figure 1 This is a flowchart of the physical constraint-guided multi-source data fusion and state completion method for power distribution networks according to the present invention.

[0031] Figure 2 This is a flowchart of the multi-source heterogeneous operation data acquisition and preprocessing process in step S1 of the present invention.

[0032] Figure 3 This is a flowchart of the construction process of the observation tensor, mask tensor, and weight tensor in step S2 of the present invention.

[0033] Figure 4 This is a flowchart of the physical constraint-guided state completion optimization model establishment process in step S3 of the present invention.

[0034] Figure 5 This is a flowchart of the iterative solution process based on the alternating direction multiplier method in step S4 of the present invention.

[0035] Figure 6 This is a flowchart of confidence assessment and bad data identification in step S5 of the present invention.

[0036] Figure 7 This is a comparison diagram of the missing and complete voltage observation process at node 4 in the illustrative example of this invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0038] For ease of description, the node set is: The set of branches is The time index set is The feature set is The observation tensor is The mask tensor is The weight tensor is The tensor of the state to be completed is The completion result is The confidence tensor is Suspicious bad data is marked as For any branch road Its resistance and reactance are respectively and ,node The set of downstream child nodes is This embodiment is applicable to radial or weak-ring distribution networks; when topology changes occur, corresponding methods can be adopted for each time segment. The physical consistency constraint terms are constructed with the branch parameters, and the method and process remain unchanged.

[0039] like Figure 1 As shown, this invention provides a method for physical constraint-guided multi-source data fusion and state completion in distribution networks, comprising the following steps:

[0040] S1: Acquisition and preprocessing of multi-source heterogeneous operational data;

[0041] S2: Construction of high-dimensional observation tensor and weight tensor;

[0042] S3: Establishment of a state completion optimization model guided by physical constraints;

[0043] S4: Iterative solution based on the alternating direction multiplier method;

[0044] S5: Confidence assessment and bad data identification.

[0045] like Figure 2 As shown, in a specific embodiment, step S1 further includes the acquisition and preprocessing of multi-source heterogeneous operating data, specifically including the following steps:

[0046] S11: Anomaly cleaning and validity assessment: Obtain raw data, including electrical measurement data and external auxiliary data, from the multi-source information system of the distribution network. The electrical measurement data should at least include node voltage amplitudes. Node-injected active power Node-injected reactive power Branch active power Branch reactive power Branch current One or more of the following: external auxiliary data includes at least one or more of meteorological data, daily type data, and equipment parameter data; physical range constraints and logical consistency checks are performed on the raw data, and records that clearly exceed the rated range of the equipment or violate basic electrical logic are judged as invalid and marked as missing. At the same time, box plot method or Hampshire filtering is used to identify isolated spikes and short-term jump outliers. Observations identified as abnormal are also marked as missing and their missing marks are retained.

[0047] S12: Feature standardization processing. To eliminate dimensional differences and improve the numerical stability of subsequent optimization, each feature is processed... Perform Z-score standardization, and let the standardized value be... ,in and It is obtained from historical windows or sliding windows and used in the output stage to de-standardize and restore it to engineering dimensions.

[0048] S13: Multi-timescale alignment, setting a uniform time step And construct a unified timeline For sampling frequencies higher than The data within each time window is taken as the mean or median as the representative value for that time step. For sampling frequencies lower than [a certain value], [the data is used to]... The data is aligned using linear interpolation or forward hold; the alignment process is only used to form a unified time index, and missing positions are still treated as missing and their labels are retained, thus obtaining a standardized multi-source dataset.

[0049] like Figure 3 As shown, in a specific embodiment, step S2 further includes the construction of the high-dimensional observation tensor and the weight tensor, specifically including the following steps:

[0050] S21: Construction of the third-order observation tensor, sorting the standardized data obtained in step S1 by index. Write observation tensor ,in The characteristic f observations representing node n at time step t

[0051] S22: Mask tensor construction, defining the mask tensor Used to identify missing locations, when There is a valid observation time. =1, when the position is missing or is determined to be invalid in step S1. This allows missing points and missing segments to be expressed uniformly under the same structure.

[0052] S23: Construction of the weight tensor. This is used to characterize the reliability differences between different data sources and different alignment qualities; in one embodiment, it is first characterized by data source type and measurement accuracy. Given basic weights Then, based on the synchronization error With packet loss rate Adjust the weights, for example, let ,in and The coefficients are non-negative; subsequently, for Normalization is performed to bring it into a stable range and facilitate numerical solution. To avoid introducing invalid constraints due to missing locations, the effective weights in the observation consistency term are adjusted. Naturally limited to The location enables unified spatiotemporal representation and reliability weighting of multi-source data.

[0053] like Figure 4 As shown, in a specific embodiment, step S3 further includes establishing a physically constrained state completion optimization model, specifically including the following steps:

[0054] S31: Definition of the state tensor to be completed, defining the entire network state tensor to be restored. Its observation position should be as close as possible to the ground. The model infers and restores reasonable values ​​at the missing locations; the completion result is denoted as... .

[0055] S32: State completion optimization model construction. The preferred objective function is a combination of low-rank prior and weighted observation consistency. A physical consistency constraint term is introduced, resulting in the following optimization problem: ,in For the tensor nuclear norm, It is the Frobenius norm. Indicates element-wise product. , , These are non-negative weighting coefficients. This is a physical consistency constraint. The low-rank constraint term is used to characterize the correlation of distribution network state in the node dimension and time dimension, so as to recover missing data by utilizing the potential low-rank structure between multiple nodes and multiple time periods. This is a weighted observation consistency term used to ensure that the completed results at existing observation locations remain consistent with the original observations. Used to define valid observation locations, This is used to represent the reliability differences corresponding to different data sources, different synchronization qualities, and different measurement accuracies. To ensure the feasibility and ease of engineering solution for the low-rank terms, The sum of the nuclear norms of the multimodal expansions can be taken as, for example ,in for The modal expansion matrix, For matrix nuclear norm, The weights are non-negative and satisfy the following conditions: .

[0056] S33: Construction of physical consistency constraints, assuming the branch set is... branch road The resistance and reactance are respectively and The set of downstream child nodes of node j is In scenarios such as radial or weak ring networks, which can be approximated by linearized models, for any given time... The residuals are constructed using the following relationship: , , Based on this, the voltage drop residual is defined. Active power balance residual Reactive power balance residual and order When partial branch power , When it cannot be directly measured, it can be incorporated as a state component. Synchronous completion, or elimination representation using linear operators, is used to ensure... Keep as about The solvable form is combined with the observation consistency term to constrain the completion process.

[0057] like Figure 5 As shown, in a specific embodiment, step S4 further includes an iterative solution based on the alternating direction multiplier method, specifically including the following steps.

[0058] S41: Variable Splitting and Augmented Lagrange Construction, Introducing Auxiliary Variables Carry low-rank terms and apply consistency constraints Simultaneously, scaling dual variables are introduced. With penalty parameters An augmented Lagrange structure is constructed to separate low-rank constraints from quadratic constraints.

[0059] S42: Low-rank approximation subproblem update, updated in the k-th iteration. : This subproblem can be solved by tensor singular value thresholding, which is equivalent to writing as ,in A thresholding operator that matches the selected tensor kernel norm.

[0060] S43: Joint update of physical correction subproblem and observation consistency, update : Under the condition that the physical residuals are expressed in a linearized form, the above equation is in the form of weighted least squares and can be solved by a sparse linear system of equations; when the scale is large, iterative methods such as conjugate gradients can be used to solve the problem to ensure efficiency and numerical stability.

[0061] S44: Dual variable update and convergence criterion, according to... Update the dual variable and use the original residuals. With dual residual Convergence is determined jointly when both are not greater than the threshold. Alternatively, iteration can stop when the maximum number of iterations is reached, yielding the completed result. And according to the standardized parameters of step S1 De-standardization is restored to engineering dimensions.

[0062] like Figure 6 As shown, in a specific embodiment, step S5 further includes confidence assessment and bad data identification, specifically including the following steps:

[0063] S51: Calculation of observation residuals and physical residuals for any index Define observation residuals Simultaneously, based on the set of branches adjacent to node n... , , Extracting local physical residual intensity It can be taken as the square root or weighted sum of the squared residuals associated with that node, to reflect the degree of deviation of the physical consistency of that time section.

[0064] S52: Confidence Tensor Generation, Definition of Confidence Tensor To characterize the reliability of the completion result, a residual mapping form is preferred, for example... ,in and The coefficient is non-negative; when When the observation residual term is zero, the confidence level is mainly reflected by the strength of the physical residual, thus avoiding misjudging missing locations as observation anomalies.

[0065] S53: Bad data labeling output, used to identify bad data based on the original observation points. And simultaneously satisfy and season Otherwise ,in and The preset threshold is used; when secondary completion is required, it can be... The location is considered an untrusted observation and the mask is updated accordingly. The solution is then recalculated to reduce the impact of hidden bad data on the complete network result. The final output includes the completed network state tensor. Confidence tensor and suspicious bad data markers It is used for subsequent status assessment, anomaly analysis, and application invocation.

[0066] To further illustrate the working mechanism and technical effectiveness of the physical constraint-guided state completion optimization model described in this invention under missing observation conditions, an illustrative example is given below. For example... Figure 7 As shown, this example constructs a 5-node radial distribution network. The node voltage amplitude, node injected active power, node injected reactive power, and branch active and reactive power are selected as state features to form a state tensor to be completed. Based on this, the artificial missing observations for some node periods are masked, and a weight tensor is constructed according to the reliability of the data source. Then, a state completion optimization model is established based on low-rank constraint terms, weighted observation consistency terms, and physical consistency constraints, and iteratively solved using the alternating direction multiplier method. Figure 7 The illustration shows that when a local missing voltage measurement occurs at node 4, the present invention can recover the trend of the missing interval and maintain the consistency between the completion result and the existing observations and the overall state evolution. This demonstrates the present invention's ability to recover locally missing segments and the physical reliability of the completion result under low observability and multi-source heterogeneity conditions.

[0067] This invention also provides a method for implementing physical constraint-guided multi-source data fusion and state completion of a distribution network using the method described above, comprising: a data acquisition and preprocessing module, a tensor construction module, a model construction module, a completion and solution module, and a result output module; wherein,

[0068] The data acquisition and preprocessing module is used to acquire multi-source heterogeneous operation data of the distribution network and perform preprocessing. The multi-source heterogeneous operation data includes at least electrical measurement data of some nodes or branches and external auxiliary data related to the electrical measurement data. The preprocessing includes at least data cleaning, time alignment and feature standardization.

[0069] The tensor construction module is used to construct observation tensors based on preprocessed multi-source heterogeneous operational data. Observation mask tensor with weight tensor ,in For the number of nodes, For time steps, For feature dimensions;

[0070] The model building module is used to establish the state tensor to be completed. The state completion optimization model includes at least a low-rank constraint term, a weighted observation consistency term, and a physical consistency constraint term.

[0071] The completion and solution module is used to iteratively solve the state completion optimization model using the alternating direction multiplier method to obtain the completed state tensor. Based on the observation consistency residuals and physical consistency residuals, a confidence tensor is generated. Marking suspicious bad data and output the , and .

[0072] This invention has the following features and effects:

[0073] Unified fusion and robust completion of multi-source heterogeneous data: Electrical measurement and external auxiliary data are organized in a unified manner by a third-order tensor, and a mask tensor and a reliability weight tensor are explicitly constructed. This enables the model to stably restore the state of the entire network even when the measurement coverage is low, communication packet loss occurs, and the quality of multi-source data fluctuates, thereby reducing the dependence on measurement redundancy and communication bandwidth.

[0074] Physical consistency-guided reliable electrical completion: A physical consistency constraint term based on linearized power flow and power balance relationship is introduced into the objective function to ensure that the completion result strictly meets the electrical laws of the distribution network. This significantly reduces the inconsistencies between voltage, power and line parameters and power balance that may occur with pure statistical methods, and improves the direct usability of the completion data in state estimation, anomaly analysis and operation assessment.

[0075] Efficient and scalable iterative solution strategy: The alternating direction multiplier method is used to split the low-rank approximation and physical correction into independent subproblems, which are solved efficiently by tensor singular value thresholding and sparse weighted least squares, respectively, taking into account both numerical stability and computational scalability, and are applicable to feeders of different sizes and long time windows.

[0076] Confidence assessment and bad data identification capabilities: Based on the joint statistics of observation consistency residuals and physical consistency residuals, the system outputs a confidence tensor and suspicious bad data labels, providing interpretable reliability information for the completion results. This supports the identification of hidden bad data and closed-loop governance of secondary completion, thereby improving the maintainability and long-term operational quality of the data infrastructure.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for multi-source data fusion and state completion in a distribution network guided by physical constraints, characterized in that, Includes the following steps: S1: Acquire multi-source heterogeneous operation data of the distribution network and perform preprocessing. The multi-source heterogeneous operation data includes at least electrical measurement data of some nodes or branches and external auxiliary data related to the electrical measurement data. The preprocessing includes at least data cleaning, time alignment and feature standardization. S2: Constructing the observation tensor based on preprocessed multi-source heterogeneous operational data Observation mask tensor with weight tensor The observation tensor The observation mask tensor is used to characterize the measurements at each observation location. The weight tensor is used to mark valid observation locations. Used to characterize the reliability of each observation, and the , , As input to the state completion optimization model in step S3; S3: Based on the observed tensor Observation mask tensor with weight tensor Establish the tensor of the state to be completed The state completion optimization model includes at least: a low-rank constraint term for characterizing the correlation of the state tensor in the node dimension and the time dimension, a weighted observation consistency term for constraining the observation location to remain consistent with the original observation, and a physical consistency constraint term for constraining the completion result to satisfy the power flow relationship and power balance relationship of the distribution network linearization. S4: The state completion optimization model is iteratively solved using the alternating direction multiplier method to obtain the completed state tensor. ; S5: Generate a confidence tensor based on observation consistency residuals and physical consistency residuals. Marking suspicious bad data and output the , and .

2. The method for multi-source data fusion and state completion in distribution networks according to claim 1, characterized in that, The electrical measurement data includes at least one of the following: node voltage amplitude. Node-injected active power Node-injected reactive power Branch active power Branch reactive power Branch current The external auxiliary data includes at least one of the following: meteorological data, daily data, and equipment parameter data; the feature dimensions It consists of electrical measurement characteristics and external auxiliary characteristics.

3. The method for multi-source data fusion and state completion in distribution networks according to claim 1, characterized in that, The preprocessing in step S1 includes: identifying outliers using box plots, Hample filtering, or Kalman filtering; and processing data from different sampling frequencies using a uniform time step. Resampling is used to achieve time alignment; linear interpolation or forward padding is used for alignment of short-time missing data, while missing location markers are preserved for constructing the observation mask tensor. .

4. The method for multi-source data fusion and state completion in distribution networks according to claim 1, characterized in that, The weight tensor The observation mask tensor is determined based on at least one of the following: data source type, measurement accuracy level, timestamp synchronization error, link latency, packet loss rate, or historical residual statistics, and then normalized. Satisfy: When In the index When there are observations at a location ,when In the index When missing .

5. The method for multi-source data fusion and state completion in distribution networks according to claim 1, characterized in that, The objective function of the state completion optimization model described in step S3 is: ； in, For low-rank constraint terms, For the weighted observation consistency term, This is a physical consistency constraint. For the tensor nuclear norm, It is the Frobenius norm. Indicates element-wise product. , , These are non-negative weighting coefficients.

6. The method for multi-source data fusion and state completion in a distribution network according to claim 5, characterized in that, The tensor null norm For the state tensor The sum of the nuclear norms after multimodal expansion, or the tensor nuclear norms based on t-SVD, can be used to characterize the correlation of the distribution network state in the node dimension and time dimension and to achieve low-rank constraints.

7. The method for multi-source data fusion and state completion in distribution networks according to claim 5, characterized in that, The physical consistency constraint Constructed based on linearized branch power flow model and power balance relationship; for any time t and any branch... ,satisfy , , ,in and Branch roads The resistance and reactance, For nodes The set of downstream child nodes; and constructed using the sum of squares of the residuals of the equation. .

8. The method for multi-source data fusion and state completion in distribution networks according to claim 1, characterized in that, The alternating direction multiplier method in step S4 includes: splitting the low-rank constraint term and the physical consistency constraint term into different subproblems for alternating updates by introducing auxiliary variables; in each iteration, performing tensor singular value thresholding updates on the low-rank subproblems, performing weighted least squares updates on the subproblems containing weighted observation consistency terms and physical consistency constraints, and stopping the iteration when the original residual and the dual residual satisfy the preset convergence condition.

9. The method for multi-source data fusion and state completion in distribution networks according to claim 1, characterized in that, The confidence tensor mentioned in step S5 The weighted observation consistency residual, physical consistency residual, and the weight tensor Jointly determined; when the residual corresponding to an observation point meets the preset criteria, the observation point is marked as suspicious bad data and written into the suspicious bad data marker. .

10. A state completion apparatus for implementing the method according to any one of claims 1 to 9, characterized in that, include: The system comprises a data acquisition and preprocessing module, a tensor construction module, a model construction module, a completion and solution module, and a result output module. The data acquisition and preprocessing module is used to acquire and preprocess multi-source heterogeneous operation data of the distribution network. The multi-source heterogeneous operation data includes at least electrical measurement data of some nodes or branches, as well as external auxiliary data related to the electrical measurement data. The preprocessing includes at least data cleaning, time alignment, and feature standardization. The tensor construction module is used to construct observation tensors based on preprocessed multi-source heterogeneous operational data. Observation mask tensor with weight tensor ,in For the number of nodes, For time steps, For feature dimension; The model building module is used to establish the state tensor to be completed. The state completion optimization model includes at least a low-rank constraint term, a weighted observation consistency term, and a physical consistency constraint term. The completion and solution module is used to iteratively solve the state completion optimization model using the alternating direction multiplier method to obtain the completed state tensor. Based on the observation consistency residuals and physical consistency residuals, a confidence tensor is generated. Marking suspicious bad data and output the , and .

Images