A power distribution network topology adaptive state estimation method and system and related device

Abstract

The application provides a power distribution network topology adaptive state estimation method and system and related devices, and belongs to the field of power system monitoring and control. The method comprises the following steps: obtaining a measurement vector and a topology state based on obtained real-time operation data of a power distribution network; taking the measurement vector and the topology state as inputs of a physical information autoencoder, and outputting voltage real part estimation values and voltage imaginary part estimation values of the power distribution network; converting the voltage real part estimation values and the voltage imaginary part estimation values into voltage amplitude estimation values and phase angle estimation values, and completing topology adaptive state estimation of the power distribution network. The physical information autoencoder comprises an input projection layer, a first residual network module, a topology perception latent module, a second residual network module and an output projection layer in sequence. The application has the advantages of high estimation accuracy, high calculation efficiency, strong physical consistency and topology change adaptability, and can effectively adapt to the operation scene of frequent topology changes of the power distribution network.

Description

Technical Field

[0001] This invention belongs to the field of power system monitoring and control, and specifically relates to a method, system and related devices for adaptive state estimation of distribution network topology. Background Technology

[0002] With the deepening of energy transition, the penetration rate of distributed energy and electric vehicles in distribution networks continues to increase. The resulting bidirectional power flows and highly variable loads bring unprecedented uncertainty and complexity to grid operation. Against this backdrop, Distribution System State Estimation (DSSE), as a core technology for real-time monitoring and optimized control, is of great significance for achieving autonomous control, demand response coordination, and protection functions. Simultaneously, modern distribution networks exhibit increasingly dynamic characteristics; frequent switching operations, fault isolation, and network reconfiguration lead to continuous changes in topology, placing higher demands on the real-time performance and adaptability of state estimation technologies.

[0003] Existing DSSE methods are mainly divided into two categories: model-driven methods and data-driven methods. Model-driven methods use weighted least squares (WLS) estimation as a basic framework and employ mathematical optimization techniques to handle measurement uncertainties and nonlinear relationships. To alleviate the observability problem caused by sparse sensor coverage, researchers have proposed auxiliary strategies such as optimizing sensor configuration, generating pseudo-measurements, and matrix completion. However, model-driven methods have two main limitations: First, relying on iterative optimization leads to a heavy computational burden. The non-convexity of the power flow equations often causes convergence problems, and changes in the distribution network topology require frequent model reconstruction and re-initialization, which limits the ability to apply in real time. Second, although topology information is usually available, distribution networks generally suffer from parameter uncertainties, including deviations in line parameters caused by factors such as equipment aging, installation differences, and changes in environmental conditions. These uncertainties significantly reduce the robustness of traditional WLS methods.

[0004] Data-driven methods, by learning the complex relationships between measurements and state variables from historical data, have shown promising application prospects. Early studies used shallow neural networks to generate voltage estimates as a warm start for Gauss-Newton iterations, improving computational efficiency. Subsequent research explored various methods, including noise-robust frameworks, geographic information integration, multi-timescale architectures, and physically guided bilinear networks embedded with Kirchhoff's laws. However, most physical information neural network methods still assume a fixed topology and lack the ability to adapt to changes in the topology of actual distribution networks.

[0005] In recent years, graph neural network (GNN)-based methods have directly embedded the power grid topology and electrical relationships into the message passing mechanism by performing graph modeling of the power distribution network topology. Although GNNs can inherently capture topological information and have the flexibility to adapt to network changes, their computational efficiency during training and inference is generally lower than that of multilayer perceptrons using vectorized inputs. This efficiency loss stems from multi-hop neighborhood aggregation operations on sparse graphs, involving irregular memory access patterns and frequent sparse matrix operations, thus limiting their scalability in real-time DSSE.

[0006] In summary, existing DSSE technology still has significant shortcomings in dealing with frequent changes in distribution network topology, parameter uncertainty, and the need for real-time calculation. There is an urgent need to develop new state estimation methods that combine high accuracy, strong adaptability, and high efficiency. Summary of the Invention

[0007] The purpose of this invention is to provide a method, system, and related apparatus for adaptive state estimation of distribution network topology. This addresses the problems of low computational efficiency, sensitivity to parameter deviations, and frequent re-initialization required by traditional model-driven methods, as well as the lack of topology adaptability and insufficient physical consistency in existing data-driven methods under conditions of frequent topology changes, widespread parameter uncertainty, and real-time operation monitoring. This invention achieves rapid and accurate estimation of different topologies by constructing a topology-aware encoder and a decoder embedded with Kirchhoff's laws, while ensuring the physical consistency of the estimation results. This significantly improves the accuracy, generalization ability, and computational efficiency of distribution network state estimation in dynamic network environments.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for adaptive state estimation of distribution network topology, comprising the following steps: Based on the acquired real-time operation data of the distribution network, measurement vectors and topology status are obtained; The measurement vector and topology state are used as inputs to the physical information autoencoder, and the outputs are the estimated real part and the estimated imaginary part of the voltage in the distribution network. The obtained voltage real part estimates and voltage imaginary part estimates are converted into voltage magnitude estimates and phase angle estimates to complete the topology adaptive state estimation of the distribution network, where: The physical information autoencoder sequentially includes an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer.

[0009] Preferably, the expression for the physical information autoencoder is:

[0010] in, In voltage state, , and These are the estimated values ​​of the real part and the estimated value of the imaginary part of the voltage, respectively. For residual network modules, It is an identity mapping; For topology-aware potential modules; For the input projection matrix; To output the projection matrix; For measurement vectors; This is the topological state; This is a composite operation for network mapping; This indicates a splicing operation.

[0011] Preferably, the expression for the topology-aware latent module is:

[0012] in, As a potential characterization, and All are linear transformation matrices. Represents the Hadamard product; It is a potential module for topology awareness.

[0013] Preferably, the expressions for the first residual network module and the second residual network module are the same, wherein the expression for the first residual network module is:

[0014] in, and This is the LeakyReLU activation function; and Both are weight matrices; and All are bias vectors; For potential characterization; This is a residual network module.

[0015] Preferably, the loss function of the physical information autoencoder is:

[0016] in, The weights for the loss function fitted to the data; The weights of the physical consistency loss function; Fit a loss function to the data; The physical consistency loss function; This is the loss function of the physical information autoencoder.

[0017] Preferably, the method for constructing the physical consistency loss function is as follows: Acquire historical operating data of the distribution network under different operating conditions, including topology status and measurement vectors; Construct a complete branch-node association matrix based on the topology information of the distribution network; Based on the topological state, and combined with the complete branch-node association matrix, a branch-node association matrix is ​​constructed; Set learnable line conductance and susceptance vectors, and calculate the node admittance matrix by combining the branch-node correlation matrix; Calculate the intermediate power vector based on the nodal admittance matrix; The intermediate power vector is projected to obtain the reconstructed measurement vector; The physical consistency loss function is constructed based on the measured measurement vectors and the reconstructed measurement vectors.

[0018] Preferably, the method for constructing the data fitting loss function is as follows: Acquire historical operating data of the distribution network under different operating conditions. The historical operating data includes topology status, measurement vector, real part of voltage, and imaginary part of voltage. The topological state and measurement vector are used as inputs to the physical information autoencoder, and the outputs are the estimates of the real part and the imaginary part of the voltage. Based on the real and imaginary parts of the voltage, as well as the estimates of the real and imaginary parts of the voltage, a data fitting loss function is constructed.

[0019] Secondly, the present invention provides a distribution network topology adaptive state estimation system, comprising: The data acquisition unit is configured to obtain measurement vectors and topology status based on the acquired real-time operating data of the distribution network. The voltage parameter acquisition unit is configured to take the measurement vector and topology state as inputs to the physical information autoencoder and output the estimated real part and estimated imaginary part of the voltage of the distribution network. The state estimation unit is configured to convert the obtained voltage real part estimate and voltage imaginary part estimate into voltage magnitude estimate and phase angle estimate, thereby completing the topology adaptive state estimation of the distribution network, wherein: The physical information autoencoder sequentially includes an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer.

[0020] Thirdly, the present invention provides an electronic device including a processor and a memory, wherein the memory stores computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the aforementioned adaptive state estimation method for power distribution network topology.

[0021] Fourthly, the present invention provides a computer program product containing computer-executable instructions, which, when executed, implement the aforementioned method for adaptive state estimation of power distribution network topology.

[0022] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a distribution network topology adaptive state estimation method. Addressing the problems mentioned in the background art, such as model-driven methods (e.g., weighted least squares) relying on iterative optimization, heavy computational burden, and frequent model reconstruction and reinitialization required for topology changes, and the low computational efficiency and difficulty in real-time application of data-driven methods (especially graph neural networks) due to multi-hop neighborhood aggregation operations on sparse graphs, this application constructs a physical information autoencoder comprising an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer. It directly uses measurement vectors and topology states as input and output voltage real and imaginary part estimates, achieving end-to-end non-iterative estimation. This vectorized fully connected architecture avoids the complex graph construction and sparse matrix operations of graph neural networks, significantly improving inference speed and adapting to the real-time monitoring needs of frequent distribution network topology changes. It also overcomes the shortcomings of traditional methods requiring remodeling during topology changes, possessing rapid response capabilities for topology adaptation.

[0023] Furthermore, by introducing an element-wise multiplication mechanism between the topology vector and the latent feature representation in the topology-aware latent module in the encoder, adaptive learning of changes in the distribution network topology is achieved. This eliminates the need to retrain or initialize the model for each topology, effectively solving the problem that traditional model-driven methods require frequent model reconstruction when the topology changes, as well as the limitation of existing data-driven methods in generalizing to unseen topologies.

[0024] Furthermore, by acquiring historical operating data of the distribution network, constructing a complete branch-node correlation matrix, dynamically generating the branch-node correlation matrix based on the topology state, setting learnable line conductance and susceptance vectors and calculating the node admittance matrix, and then reconstructing the intermediate power vector based on the bilinear power flow formula and projecting it to obtain the reconstructed measurement vector, a physical consistency loss function is finally constructed based on the measured and reconstructed measurement vectors. The basic physical constraints of the power system, such as Kirchhoff's laws and Ohm's laws, are explicitly embedded into the training process of the decoder. This method enables the decoder to adaptively fit the actual line parameters through learnable parameters without relying on accurate prior information of line parameters. It effectively addresses the parameter uncertainty caused by factors such as equipment aging and environmental changes, while ensuring that the estimation results meet physical laws, thus enhancing the robustness and interpretability of the model. Attached Figure Description

[0025] Figure 1 This is a flowchart of an embodiment of the present invention; Figure 2 This is a system block diagram provided for an embodiment of the present invention; Figure 3 This is an overall framework diagram of the physical information autoencoder according to an embodiment of the present invention; Figure 4 This is a detailed architecture diagram of the physical information autoencoder according to an embodiment of the present invention. Detailed Implementation

[0026] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0027] It should be understood that, when used in this application specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0028] It should also be understood that the term “and / or” as used in this application specification means any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0029] As used in this application specification, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."

[0030] Furthermore, in the description of this application, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0031] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0032] Example 1 This embodiment provides a method for adaptive state estimation of distribution network topology, including the following steps: Based on the acquired real-time operation data of the distribution network, measurement vectors and topology status are obtained; The measurement vector and topology state are used as inputs to the physical information autoencoder, and the outputs are the estimated real part and the estimated imaginary part of the voltage in the distribution network. The obtained voltage real part estimates and voltage imaginary part estimates are converted into voltage magnitude estimates and phase angle estimates to complete the topology adaptive state estimation of the distribution network, where: The physical information autoencoder sequentially includes an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer.

[0033] Example 2 like Figures 1 to 4 As shown in the figure, this embodiment provides a distribution network topology adaptive state estimation method, which includes the following steps: Step 1: Collect historical operating data of the distribution network, including measurement data, topology information and voltage status under different operating conditions; Step S101: Collect historical operating data of the distribution network under different operating conditions, and preprocess it into the following format: (1) Topology of the distribution network ,in, Indicating the distribution network The connection status of the line, when Time indicates line Connection, when Time indicates line disconnect, This represents the total number of distribution network lines. It is a real number.

[0034] (2) Measurement vector This includes node power injection and branch power flow measurement. The total number of measurements; (3) Voltage status, including node voltage amplitude and phase angle , The total number of distribution network nodes, and through , Converted to the real part of voltage and the virtual part spare; Step S102: Based on the topology information of the distribution network, construct a complete branch-node association matrix for all candidate lines, including backup lines. For connecting nodes and nodes branch road The corresponding elements in the complete branch-node association matrix satisfy the following: , The remaining elements are 0.

[0035] Step 2: Construct a physical information autoencoder, which consists of an encoder and a decoder; Step S201, Build the encoder architecture: Input layer: Receives measurement vectors and topological state And then splice them together; Input projection layer: via input projection matrix Mapping the input to a latent space of a specific dimension; The first residual network module performs deep feature learning on the projected features to obtain the latent representation; Topology-aware latent modules capture topological features through element-wise product of topological states and latent representations; The second residual network module performs further deep feature learning on the topology-aware features; Output projection layer: through the output projection matrix The features output by the second residual network module are mapped to the output dimension, namely the real and imaginary parts of the output voltage. The overall mapping relationship of the encoder is represented as follows:

[0036] in, In voltage state, , and These are the estimated values ​​of the real part and the estimated value of the imaginary part of the voltage, respectively. For residual network modules, It is an identity mapping; For topology-aware potential modules; Indicates a splicing operation; The composite operation representing network mapping is used to characterize the cascade relationship of modules in the network according to the forward propagation order; Among them, topology-aware latent modules The expression is:

[0037] in, As a potential characterization, and All are linear transformation matrices. Represents the Hadamard product; This is the output of the topology-aware latent module.

[0038] Among them, residual network module The expression is:

[0039] in, and All are LeakyReLU activation functions. and Both are weight matrices. and All are bias vectors; This is the output of the residual network module.

[0040] Step S202, construct the decoder architecture: Construct a branch-node association matrix related to the topology based on the topology state:

[0041] in, To be based on topological state A diagonal matrix with diagonal elements. The complete branch-node association matrix; Indicates the state of a given topology The branch-node correlation matrix obtained under the action; Based on the set learnable line conductivity and susceptance vector Calculate the nodal admittance matrix:

[0042]

[0043] in, Here is the conductance matrix. The susceptance matrix; Based on the bilinear power flow formula, the nodal active power injection and nodal reactive power injection are reconstructed:

[0044]

[0045] in, and These refer to active power injection and reactive power injection at the nodes, respectively. It is a vector consisting entirely of 1s; This is the matrix transpose.

[0046] Reconstructing the active and reactive power flows at both ends of the branch road, for the branch road ends ,have: in, and Branch ends The trend of meritorious and ineffective actions , , , ; and These are the branch-node association matrices corresponding to the beginning and end points of the branch, respectively, and their matrix dimensions are... The same, among which, The value is 1 only at the location where the branch connects to the sending bus. The value is 1 only at the point where the branch connects to the receiving busbar; all other elements are 0. This is a vector-to-diagonal matrix conversion operation used to generate a diagonal matrix with the vector as the main diagonal element.

[0047] Concatenate all the reconstructed power values ​​into an intermediate power vector. :

[0048] in, and These are node active power injection and node reactive power injection, respectively. For the active power flow at the end of the branch line; For reactive power flow at the end of the branch; The meritorious current at the head of the branch road; For the reactive power flow at the beginning of the branch road; This is the matrix transpose.

[0049] The intermediate power vector is projected onto the observation space to obtain the reconstructed measurement. :

[0050] in, A measurement selection matrix is ​​used to select the actual configured measurement from all power quantities; This is the intermediate power vector.

[0051] Step 3: Use a two-stage training strategy to jointly optimize and train the physical information autoencoder; Step S301: Fix the encoder parameters and input the actual real part of the voltage to the decoder. virtual part and topological state The reconstructed measurements were obtained. Using the physical consistency loss function Training the decoder:

[0052] in, and The first The actual value of the measurement and the first One measurement reconstructed value, For the first The variance of the measurement error; This represents the total number of measurements.

[0053] Step S302: Fix the decoder parameters and construct the data fitting loss function. :

[0054] in, The number of distribution network nodes, and These are the true values ​​of the real and imaginary parts of the voltage, respectively. and These are the estimated real part and estimated imaginary part of the voltage output by the encoder, respectively. This is the balance coefficient between the real and imaginary part losses; Construct a comprehensive loss function :

[0055] in, and The weights for the two types of losses; Fit a loss function to the data; This is the physical consistency loss function.

[0056] Initialize weight coefficients and set weights. This ensures that the initial real and imaginary losses are numerically balanced; setting and This ensures that the initial data loss and the initial physical loss are numerically balanced; Using the comprehensive loss function Train the encoder.

[0057] Step 4: Use the trained model to perform real-time state estimation of the distribution network.

[0058] Step S401: Obtain real-time operation data of the distribution network and preprocess it into measurement vectors. and topological state ; Step S402: Input the real-time running measurements and topology into the trained encoder to obtain the estimated real part and estimated imaginary part of the voltage. and ; In step S403, optionally, the voltage estimate is input into the trained decoder to obtain the predicted power flow. ; Step S404: Combine the estimated real part of the voltage with the estimated imaginary part of the voltage. and The voltage amplitude estimate can be converted using the following formula. and phase angle estimates :

[0059]

[0060] Example 3 This embodiment uses the IEEE standard test system to verify the proposed adaptive state estimation method for distribution network topology based on physical information autoencoders. The specific steps are as follows: (1) Experimental setup This embodiment selects IEEE distribution networks with 33, 69, 84, 118, 136, and 874 nodes as test objects. Each test system adopts a radial operating configuration, and its specific configuration parameters are shown in Table 1. Photovoltaic generators are deployed at 10% of the system nodes, and the distributed energy penetration rate is set to 20%. Load consumption and photovoltaic power generation curves are scaled based on real data.

[0061] Table 1 Test System Configuration Parameters

[0062] For the measurement configuration, real-time measurements include active and reactive power injection measurements at all nodes except the balancing node; and active and reactive power flow measurements at 20% of the system branches. All measurements follow a Gaussian noise distribution with a standard deviation of 1%.

[0063] To evaluate the robustness and adaptability of the method under different operating conditions, datasets containing various network topologies were generated, as shown in Table 2. The training dataset uses single-pair switch operations, i.e., closing one switch and opening another from the basic topology. The T1 test set follows the same distribution as the training data to evaluate performance within the distribution; the T2 and T3 test sets introduce multi-branch topology variations by including 2-pair and 3-pair switch scenarios, gradually increasing the complexity beyond the single-pair switch cases in the training set.

[0064] Table 2 Number of topologies in the training and testing datasets

[0065] To evaluate the effectiveness of the proposed method, it is compared with several representative methods, including the traditional weighted least squares (WLS) method and four data-driven methods. The data-driven methods include a multilayer perceptron (MLP) with three hidden layers, and three graph neural network (GNN) methods: graph attention network (GAT), XENet, and PowerFlowNet (PFN). Furthermore, all GNN models are configured with a three-layer GNN network structure.

[0066] The evaluation metrics are set as Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE):

[0067]

[0068] in, For the number of test samples, Indicates the true value of the voltage state. This represents the estimated voltage state value.

[0069] (2) Physical consistency assessment The decoder network is trained to estimate line parameters using the method in step S301 of Example 2. Experimental results are shown in Table 3. The method demonstrates excellent accuracy across all test systems, with MAPE estimates for conductance ranging from 0.02% to 3.31% and for susceptance ranging from 0.16% to 3.90%. Therefore, this method exhibits strong robustness to parameter uncertainties.

[0070] Table 3 shows the MAPE (Mean Average Parameter Estimation) of the decoder for branch parameters.

[0071] (3) State estimation results After implementing all the steps of Example 2, the state estimation results for the T1 test set are shown in Table 4. This method achieves optimal state estimation accuracy in all test systems. Voltage amplitude MAPE as low as Voltage phase angle MAE as low as The accuracy of this method is significantly better than that of WLS and other data-driven methods. Therefore, this method has excellent estimation accuracy and strong adaptability to topological changes.

[0072] Table 4 shows the state estimation results of each method on the T1 test set, which is distributed similarly to the training set.

[0073] Note: MAPE is used for voltage amplitude. The unit is MAE is used for voltage phase angle. The unit is Degree. Bold values ​​indicate the optimal performance result under each evaluation metric.

[0074] (4) Generalization ability analysis To evaluate the generalization ability of this method under network topology changes, comprehensive experiments were conducted on the T2 and T3 test sets. The experimental results are shown in Tables 5 and 6. The proposed method achieves accuracy comparable to WLS on the T2 test set under moderate topology changes and consistently outperforms other data-driven methods in all test scenarios. In scenarios with severe topology changes, although the accuracy is slightly lower than WLS, this method still maintains high estimation accuracy and significantly improves computational speed. Therefore, this method demonstrates excellent generalization ability and accuracy under topology changes.

[0075] Table 5. State estimation results for each method on the T2 test set.

[0076] Note: MAPE is used for voltage amplitude. The unit is MAE is used for voltage phase angle. The unit is Degree. Bold and underlined values ​​represent optimal and suboptimal performance, respectively.

[0077] Table 6. State estimation results for each method on the T3 test set.

[0078] Note: MAPE is used for voltage amplitude. The unit is MAE is used for voltage phase angle. The unit is Degree. Bold and underlined values ​​represent optimal and suboptimal performance, respectively.

[0079] (5) Evaluation of computational efficiency The computational efficiency of this method was evaluated based on the training and testing times of the model, and the results are shown in Tables 7 and 8. As shown in Table 7, this method and MLP exhibit excellent scalability; the training time remains almost constant (3-8 seconds per round) as the distribution network scales up, while GNN-based methods show significant degradation, with the training time for PFN increasing sharply to 42 seconds per round for a 136-node system. As shown in Table 8, this method achieves the fastest estimation performance in all test scenarios, with an average computation time per sample of 0.010 to 0.031 milliseconds, comparable to MLP performance and significantly faster than WLS and various GNN methods. Therefore, this method has excellent computational efficiency.

[0080] Table 7. Average training time per round for each data-driven method (unit: seconds)

[0081] Table 8. Average test time per sample for each method (unit: milliseconds)

[0082] (6) Testing on large-scale systems The state estimation performance on a large-scale 874-node distribution network is shown in Table 9. In the 874-node system, our method achieves the highest accuracy on the distributed test set, maintains robust performance under topology change scenarios, and reduces the inference time to 0.040 milliseconds, achieving an acceleration of approximately 15,000 times compared to WLS.

[0083] Table 9. State estimation results of the 874-node test system under different test scenarios.

[0084] Note: MAPE is used for voltage amplitude. The unit is MAE is used for voltage phase angle. The unit is Spend.

[0085] Example 4 This embodiment provides a distribution network topology adaptive state estimation system, including: The data acquisition unit is configured to obtain measurement vectors and topology status based on the acquired real-time operating data of the distribution network. The voltage parameter acquisition unit is configured to take the measurement vector and topology state as inputs to the physical information autoencoder and output the estimated real part and estimated imaginary part of the voltage of the distribution network. The state estimation unit is configured to convert the obtained voltage real part estimate and voltage imaginary part estimate into voltage magnitude estimate and phase angle estimate, thereby completing the topology adaptive state estimation of the distribution network, wherein: The physical information autoencoder sequentially includes an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer.

[0086] Example 5 This embodiment also provides a computing device. The computing device includes a bus, a processor, a memory, and a communication interface. The processor, memory, and communication interface communicate with each other via the bus. The computing device can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memory in the computing device.

[0087] A bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, a bus can include a path for transmitting information between various components of a computing device (e.g., memory, processor, communication interfaces).

[0088] The processor may include any one or more of the following: central processing unit (CPU), graphics processing unit (GPU), tensor processing unit (TPU), application specific integrated circuit (ASIC), field-programmable gate array (FPGA), microprocessor (MP), or digital signal processor (DSP).

[0089] Memory can include volatile memory, such as random access memory (RAM). Processors can also include non-volatile memory. volatile memory, such as read-only memory (ROM). ROM (memory only), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0090] The memory stores executable program code, which the processor executes to implement the functions of the aforementioned units, thereby achieving, for example, the method described in Embodiment 1. That is, the memory may store instructions for the methods and functions relating to the computing device in any of the above embodiments.

[0091] The communication interface uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between computing devices and other devices or communication networks.

[0092] Example 6 This embodiment also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, cause the processor to perform the methods and functions of the computing device involved in any of the above embodiments.

[0093] Generally, the various embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software, which can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or represented using some other illustration, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0094] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for adaptive state estimation of distribution network topology, characterized in that, Includes the following steps: Based on the acquired real-time operation data of the distribution network, measurement vectors and topology status are obtained; The measurement vector and topology state are used as inputs to the physical information autoencoder, and the outputs are the estimated real part and the estimated imaginary part of the voltage in the distribution network. The obtained voltage real part estimates and voltage imaginary part estimates are converted into voltage magnitude estimates and phase angle estimates to complete the topology adaptive state estimation of the distribution network, where: The physical information autoencoder sequentially includes an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer.

2. The adaptive state estimation method for distribution network topology according to claim 1, characterized in that, The expression for the physical information autoencoder is: in, In voltage state, , and These are the estimated values ​​of the real part and the estimated value of the imaginary part of the voltage, respectively. For residual network modules, It is an identity mapping; For topology-aware potential modules; For the input projection matrix; To output the projection matrix; For measurement vectors; This is the topological state; This is a composite operation for network mapping; This indicates a splicing operation.

3. The adaptive state estimation method for distribution network topology according to claim 1, characterized in that, The expression for the topology-aware latent module is: in, For potential characterization; and All are linear transformation matrices; Represents the Hadamard product; This is the topological state; This is the output of the topology-aware latent module.

4. The adaptive state estimation method for distribution network topology according to claim 1, characterized in that, The expressions for the first residual network module and the second residual network module are the same, wherein the expression for the first residual network module is: in, and This is the LeakyReLU activation function; and Both are weight matrices; and All are bias vectors; For potential characterization; This is the output of the residual network module.

5. The adaptive state estimation method for distribution network topology according to claim 1, characterized in that, The loss function of the physical information autoencoder is: in, The weights for the loss function fitted to the data; The weights of the physical consistency loss function; Fit a loss function to the data; The physical consistency loss function; This is the loss function of the physical information autoencoder.

6. The adaptive state estimation method for distribution network topology according to claim 5, characterized in that, The method for constructing the physical consistency loss function is as follows: Acquire historical operating data of the distribution network under different operating conditions, including topology status and measurement vectors; Construct a complete branch-node association matrix based on the topology information of the distribution network; Based on the topological state, and combined with the complete branch-node association matrix, a branch-node association matrix is ​​constructed; Set learnable line conductance and susceptance vectors, and calculate the node admittance matrix by combining the branch-node correlation matrix; Calculate the intermediate power vector based on the nodal admittance matrix; The intermediate power vector is projected to obtain the reconstructed measurement vector; The physical consistency loss function is constructed based on the measured measurement vectors and the reconstructed measurement vectors.

7. The adaptive state estimation method for distribution network topology according to claim 5, characterized in that, The method for constructing the data fitting loss function is as follows: Acquire historical operating data of the distribution network under different operating conditions. The historical operating data includes topology status, measurement vector, real part of voltage, and imaginary part of voltage. The topological state and measurement vector are used as inputs to the physical information autoencoder, and the outputs are the estimates of the real part and the imaginary part of the voltage. Based on the real and imaginary parts of the voltage, as well as the estimates of the real and imaginary parts of the voltage, a data fitting loss function is constructed.

8. A distribution network topology adaptive state estimation system, characterized in that, include: The data acquisition unit is configured to obtain measurement vectors and topology status based on the acquired real-time operating data of the distribution network. The voltage parameter acquisition unit is configured to take the measurement vector and topology state as inputs to the physical information autoencoder and output the estimated real part and estimated imaginary part of the voltage of the distribution network. The state estimation unit is configured to convert the obtained voltage real part estimate and voltage imaginary part estimate into voltage magnitude estimate and phase angle estimate, thereby completing the topology adaptive state estimation of the distribution network, wherein: The physical information autoencoder sequentially includes an input projection layer, a first residual network module, a topology-aware latent module, a second residual network module, and an output projection layer.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer instructions, which, when executed by the processor, cause the electronic device to perform a power distribution network topology adaptive state estimation method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product contains computer-executable instructions, which, when executed, implement a power distribution network topology adaptive state estimation method according to any one of claims 1 to 7.

Images