An auto-encoder static power distribution network state estimation method and system

Abstract

The application relates to a self-encoder static power distribution network state estimation method and system, and belongs to the field of power systems. The method collects simulation data of a static power distribution network system to generate a repair model when part of the data is missing; based on the repair model, a target domain model is obtained through transfer learning fine-tuning; the measured data with part of the data missing is repaired to obtain complete data through the repair model, and the target domain model is used for state estimation of the static power distribution network system. The application can accurately predict the operation state of a power system.

Description

Technical Field

[0001] The present invention belongs to the field of power systems, and in particular relates to a method and system for estimating the state of an autoencoder static distribution network. Background Art

[0002] Power system state estimation is the foundation of power system monitoring and dispatch, providing critical real-time information for the safe and stable operation of power systems. Therefore, power system state estimation has always been an important area of ​​power system research. Traditional power system state estimation methods mainly include least squares estimation, weighted least squares, and Kalman filtering. Least squares estimation requires accurate and error-free measurement data, but it is prone to estimation errors in the presence of noise or data loss, resulting in inaccurate state estimation results. Weighted least squares improves noise immunity by introducing weights, but the selection of weights relies on prior knowledge, making it difficult to cope with poor data quality under actual operating conditions. Kalman filtering is effective in estimating the state of dynamic power systems, but it relies heavily on the initial state and is easily affected by external interference, resulting in large estimation errors.

[0003] Furthermore, as power systems expand in size and complexity, traditional state estimation methods face challenges such as high computational complexity and difficulty in data synchronization. In modern smart grids, power system data sources are more diverse and data volumes are significantly increased. Sensors may experience noise, data loss, and latency, making it difficult for traditional state estimation methods to meet real-time and accuracy requirements. Especially in the context of large-scale distributed energy access and the fusion of multi-source heterogeneous data, the requirements for data quality in power system state estimation have become even more stringent. Achieving efficient and accurate state estimation in the presence of noise, data loss, and latency has become a pressing issue. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for estimating the state of a static distribution network using an autoencoder in response to the existing technical problems.

[0005] To achieve the above object, the technical solution of the present invention is: a method for estimating the state of a static distribution network using an autoencoder, comprising:

[0006] When part of the input data is missing, the static distribution network system simulation model is input for simulation, the simulation data of the static distribution network system simulation model is collected, and a repair model is generated based on the simulation data of the static distribution network system simulation model;

[0007] Based on the repair model, the target domain model is obtained through transfer learning fine-tuning;

[0008] Based on the actual input data with some missing data in the actual static distribution network system, the repair model is used to repair the complete input data. Based on the complete input data, the target domain model is used to estimate the operating status of the actual static distribution network system.

[0009] In one embodiment of the present invention, partial data loss includes setting both active power P and reactive power Q in the input data to zero.

[0010] In one embodiment of the present invention, the repair model is generated as follows:

[0011] Within the framework of federated learning, the simulation data of the static distribution network system is clustered using the Gaussian mixture model (GMM), and the simulation data of the static distribution network system is divided into multiple subsets. The data of each subset is assigned to a different federated learning client, and each federated learning client independently trains an autoencoder model. The weights of the trained autoencoder model are sent to the central server, which aggregates the weights of the autoencoder models of all federated learning clients through weighted averaging to generate a repair model.

[0012] In one embodiment of the present invention, the encoder of the autoencoder model consists of four fully connected layers, which gradually compresses the dimension of the input data to a low-dimensional representation; the decoder of the autoencoder model reconstructs reconstructed data with the same dimension as the input data by expanding the compressed low-dimensional representation layer by layer.

[0013] In one embodiment of the present invention, the training method of the autoencoder model is as follows:

[0014] The encoder of the autoencoder model converts the input data x into hidden variables z, which can be expressed as:

[0015] z=f(w (1) x+b (1) )

[0016] Where w (1) represents the weight matrix; b (1) represents the bias vector; f() represents the activation function;

[0017] The decoder of the autoencoder model reconstructs data that is similar to the input data x based on the hidden variable z , which can be expressed as:

[0018]

[0019] Where w (2) represents the weight matrix; b (2) represents the bias vector;

[0020] The optimization goal of the autoencoder model is to ensure that the reconstructed data As close as possible to the input data x; given a set of input data x (n) ∈R (D) ,1≤n≤N, the reconstruction error is expressed as:

[0021]

[0022] Where, Represents reconstructed data;

[0023] By minimizing the reconstruction error, the network parameters of the autoencoder model are learned as θ = {w (1) ,b (1) ,w (2) ,b (2)}.

[0024] In one embodiment of the present invention, the activation function adopts the ReLU function:

[0025] f(Z)=max(0,Z)

[0026] Z represents input.

[0027] In one embodiment of the present invention, the central server aggregates the weights of the autoencoder models of all federated learning clients by weighted averaging as follows:

[0028] Suppose there are K federated learning clients, whose local data are represented as D1, D2, ..., D k ,…,D K , for the dataset D of the kth federated learning client k , the collected loss function is:

[0029]

[0030] Where, f j (w) is f(x j ,y j ), x j represents the input variable of the jth sample; y j represents the output variable of the jth sample; the loss function defined on the dataset set of all federated learning clients is:

[0031]

[0032] Where, |D k | indicates D k The number of samples in |D| represents the total number of samples in all federated learning clients.

[0033] The global aggregation parameter set w(t) of the central server at time t is expressed as follows:

[0034]

[0035] Where μ is the learning rate; ▽F(·) is the local gradient of the loss function; w k (t) is the local parameter set of the kth federated learning client at time t, expressed as:

[0036] w k (t) = w k (t-1)-μ▽F k (w k (t-1))

[0037] The central server obtains the local gradient ▽F of the loss function of the kth federated learning client at time t k (w k (t-1)), and w(t) is calculated.

[0038] In one embodiment of the present invention, based on the repair model, the target domain model is obtained through transfer learning fine-tuning in the following specific implementation manner:

[0039] In transfer learning, let subscripts S and T represent the source task and target task respectively, D S and D T They are the source domain and the target domain respectively, and the corresponding domains are constructed using the feature vector space X and the probability distribution function P(X), namely:

[0040]

[0041] Let Y s and Y t are the label vector spaces of the source domain and the target domain respectively, f s and f t are the mapping functions of the source domain and the target domain respectively, then the tasks T in the source domain and the target domain are s and T t They are described as:

[0042]

[0043] Assuming that the source domain and the target domain have the same feature distribution but different label spaces, the goal of transfer learning is to learn the mapping function f t :X t →X s In D T The expected error on X is the smallest and satisfies X t =X s 、Y t =Y s and P(Y t |X t )≠P(Y s |Xs );

[0044] The repair model is used as the source domain model. Through transfer learning, the feature representation and parameters of the repair model are transferred to the target domain model. The parameters of the target domain model are fine-tuned to adapt the parameters of the target domain model to the characteristics of the target domain data. The fine-tuned target domain model is then trained on the target domain data to obtain the final target domain model. The specific process is as follows:

[0045] (1) Source domain data collection: collecting node voltage (v), phase angle (θ), active power (p) and reactive power (q) data;

[0046] (2) Source domain model training: A high-precision state estimation model is trained on source domain data. This model usually uses a large amount of labeled source domain data and has good generalization capabilities. The goal of the source domain model is to capture the characteristics and change patterns of the source domain power system state data as much as possible;

[0047] (3) Feature representation transfer: transfer the feature representation of the source domain model (i.e., the extracted intermediate layer features or feature space) to the target domain model;

[0048] (4) Parameter migration and fine-tuning: Migrate some or all parameters of the source domain model to the target domain model. By fine-tuning on the target domain data, further adjust the migrated parameters to adapt the model to the characteristics of the target domain.

[0049] (5) Target domain model training: Continue to train the model on the target domain data to adapt to the state estimation task in the target domain. Through target domain training, the model can better capture the state characteristics of the target domain power system and have higher estimation accuracy;

[0050] (6) State Estimation: Use the fine-tuned transfer learning model to estimate the state of the power system in the target domain. Predict the voltage, frequency, power, and other states of each node in the system and identify potential anomalies or instabilities. Evaluate the estimation accuracy of the model to ensure that the state estimation results are sufficiently accurate and robust in the target domain.

[0051] The present invention also provides an autoencoder static distribution network state estimation system based on federated learning and transfer learning, comprising:

[0052] The simulation data acquisition module inputs the static distribution network system simulation model for simulation when part of the input data is missing, and collects simulation data of the static distribution network system simulation model;

[0053] A repair model generation module generates a repair model based on simulation data of a static distribution network system simulation model;

[0054] The transfer learning module obtains the target domain model through transfer learning fine-tuning based on the repair model;

[0055] The state estimation module uses the repair model to repair the actual input data with some missing data in the actual static distribution network system to obtain complete input data. Based on the complete input data, the target domain model is used to estimate the operating state of the actual static distribution network system.

[0056] The present invention also provides a computer-readable storage medium on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) The present invention uses the improved autoencoder structure of the federated learning framework to train a repair model, and the data repair results are more accurate.

[0059] (2) The present invention obtains the target domain model for state estimation through transfer learning fine-tuning, which does not require retraining the model and is more efficient.

[0060] (3) The present invention takes into account the data missing problem that occurs under weak communication conditions, preprocesses the measurement data, including data cleaning, completion, calibration, etc., obtains the statistical distribution characteristics of the data, and proposes a corresponding repair model. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Flow chart of the method of the present invention.

[0062] Figure 2 The results of patching are shown for P (active power) and Q (reactive power) data.

[0063] Figure 3 Estimated effects in the presence of missing data are presented. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0065] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0066] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0067] In power system state estimation, more robust and intelligent algorithms are needed that can effectively address issues such as noise, data loss, and delays, thereby improving the reliability and accuracy of state estimation. Furthermore, with the development of artificial intelligence and machine learning technologies, data-driven intelligent state estimation methods, such as deep learning, autoencoders, and federated learning, are becoming a potential solution to these problems, providing new technical approaches and solutions for power system state estimation.

[0068] To this end, the present invention provides a method for estimating the state of a static distribution network using an autoencoder, comprising:

[0069] When part of the input data is missing, the static distribution network system simulation model is input for simulation, the simulation data of the static distribution network system simulation model is collected, and a repair model is generated based on the simulation data of the static distribution network system simulation model;

[0070] Based on the repair model, the target domain model is obtained through transfer learning fine-tuning;

[0071] Based on the actual input data with some missing data in the actual static distribution network system, the repair model is used to repair the complete input data. Based on the complete input data, the target domain model is used to estimate the operating status of the actual static distribution network system.

[0072] The following is a specific implementation process of the present invention.

[0073] like Figure 1 As shown, this embodiment provides a method for estimating the state of a static distribution network from an autoencoder, such as Figure 1 As shown, the following steps are included:

[0074] S1, data acquisition, simulates the collection of simulation data of the static distribution network system under the condition of partial data missing;

[0075] S2, model construction, in the framework of federated learning, the simulation data is clustered by Gaussian Mixture Model (GMM), and the simulation data is divided into multiple subsets with strong homogeneity. The data of each subset is assigned to a different federated learning client, and each client independently trains an autoencoder model on its assigned data subset and sends the trained model weight to the central server. The central server aggregates all client model weights by weighted averaging to generate a repair model.

[0076] For input data x, the encoding process of the autoencoder can be represented by the formula:

[0077] z = f(w (1) x + b (1) )

[0078] In the formula, x represents the input data; z represents the hidden variable; w (1) represents the weight matrix; b (1) represents the bias vector; f() represents the activation function, usually using the ReLU function:

[0079] f(z) = max(0, z)

[0080] Through this mapping, the encoder converts the high-dimensional input data x into a low-dimensional latent representation z, which contains the main features of the input data.

[0081] The decoding process can be represented by the formula:

[0082]

[0083] In the formula, represents the reconstructed data; z represents the hidden variable; w (2) represents the weight matrix; b (2) represents the bias vector; f() represents the activation function, also using the ReLU function.

[0084] The autoencoder converts the input data into hidden variables z, and through the decoder, it reconstructs the reconstructed data that is as close to the original input as possible. Therefore, ensuring that the output of the decoder can recover the original data as much as possible is the optimization goal of the autoencoder, that is, the reconstructed data is as close to the input data x as possible. Given a set of data x (n) ∈R (D) , 1 ≤ n ≤ N, then the reconstruction error can be represented as:

[0085]

[0086] In the formula, x (n) represents the input data; Reconstructed data.

[0087] The most typical reconstruction error indicator is usually measured by mean square error (MSE):

[0088]

[0089] By minimizing the reconstruction error, the network parameters θ={w (1) ,b (1) ,w (2) ,b (2)}.

[0090] The autoencoder architecture designed in this paper takes into account the characteristics of power system data. The encoder, consisting of four fully connected layers, gradually compresses the input data from 81×16 dimensions to a compact low-dimensional representation. The decoder then expands this low-dimensional representation layer by layer to reconstruct a data matrix with the same dimensions as the original input.

[0091] Suppose there are K federated learning clients, whose local data are represented as D1, D2, ..., D k ,…,D K , for the dataset D of the kth federated learning client k , the collected loss function is:

[0092]

[0093] Where, f j (w) is f(x j ,y j ), x j represents the input variable of the jth sample; y j represents the output variable of the jth sample; the loss function defined on the dataset set of all federated learning clients is:

[0094]

[0095] Where, |D k | indicates D k The number of samples in |D| represents the total number of samples in all federated learning clients. The repair model is trained to minimize the loss function, for example using gradient descent, to find the optimal set of parameters to learn the objective of minimizing the loss function F k (w):

[0096] w * =argminF k (w)

[0097] Since the original data is distributed in different clients, federated learning cannot directly find the gradient on the central server as centralized machine learning. If the gradient descent method is used to minimize the global loss function, that is:

[0098]

[0099] In the formula, μ is a small positive number, that is, the learning rate; ▽F(·) is the local gradient of the loss function; w(t) is the global aggregated parameter set of the central server at time t; w k (t) is the local parameter set of the client k at time t, which can be expressed as:

[0100] w k (t)=w k (t-1)-μ▽F k (w k (t-1))

[0101] After that, as long as the local gradient ▽F k (w k (t-1)) is obtained, the central server can calculate w(t). Therefore, only the local gradient needs to be sent to the central server, which can save communication resources, especially when using gradient compression, which can reduce the amount of gradient data transmitted to a certain extent;

[0102] S3, model migration, migrate the repair model obtained by training in S2 to the target domain, fine-tune it using the data of the target domain, retain important features in the source domain, and adapt to new data features in the target domain to obtain a target domain model.

[0103] In transfer learning, let the subscripts S and T represent the source task and the target task respectively, D S and D T are the source domain and the target domain, and the feature vector space X and the probability distribution function P(X) are used to form the corresponding fields, that is:

[0104]

[0105] Let Y s and Y t be the label vector space of the source domain and the target domain, f s and f t be the mapping functions of the source domain and the target domain, and T s and T t be the tasks in the source domain and the target domain, respectively, which can be described as:

[0106]

[0107] Assuming that the source domain and the target domain have the same feature distribution but different label spaces, the goal of transfer learning is to learn the mapping function f t :X t →X s In D T The expected error on X is the smallest and satisfies X t =X s 、Y t =Y s and P(Y t |X t )≠P(Y s |X s );

[0108] Repair the model and fine-tune it through transfer learning to obtain the target domain model. The specific process is as follows:

[0109] (1) Source domain data collection: collecting node voltage (v), phase angle (θ), active power (p) and reactive power (q) data;

[0110] (2) Source domain model training: A high-precision state estimation model is trained on source domain data. This model usually uses a large amount of labeled source domain data and has good generalization capabilities. The goal of the source domain model is to capture the characteristics and change patterns of the source domain power system state data as much as possible;

[0111] (3) Feature representation transfer: transfer the feature representation of the source domain model (i.e., the extracted intermediate layer features or feature space) to the target domain model;

[0112] (4) Parameter migration and fine-tuning: Migrate some or all parameters of the source domain model to the target domain model. By fine-tuning on the target domain data, further adjust the migrated parameters to adapt the model to the characteristics of the target domain.

[0113] (5) Target domain model training: Continue to train the model on the target domain data to adapt to the state estimation task in the target domain. Through target domain training, the model can better capture the state characteristics of the target domain power system and have higher estimation accuracy;

[0114] (6) State Estimation: Use the fine-tuned transfer learning model to estimate the state of the power system in the target domain. Predict the voltage, frequency, power, and other states of each node in the system and identify potential anomalies or instabilities. Evaluate the estimation accuracy of the model to ensure that the state estimation results are sufficiently accurate and robust in the target domain.

[0115] S4, state estimation, repairs the measured data with some missing data through the repair model, and then uses the repaired data to perform state estimation on the target domain data through the fine-tuned model to ensure that the model can accurately predict the operating status of the power system.

[0116] The proposed method first sets the P and Q components to zero to generate system simulation data. Then, within the federated learning framework, the source domain data is clustered using a Gaussian mixture model (GMM). Using the expectation-maximization (EM) algorithm, the GMM partitions the source domain data into multiple, highly homogeneous subsets. The data from each subset is then distributed to different federated learning clients.

[0117] Each client independently trains an autoencoder model on its assigned data subset. Using an encoder-decoder structure, the autoencoder model extracts low-dimensional feature representations from the input data and attempts to reconstruct the input data. After each round of training, each client sends the trained model weights to a central server.

[0118] The central server aggregates the model weights of all clients through weighted averaging to generate a repair model. This repair model combines the learning results of different clients and has the ability to generalize in multiple scenarios.

[0119] Data inpainting is performed directly using the federated learning repair model. For target domain data with missing P and Q data, this data is directly input into the global model, and the decoder of the model generates the complete, inpainted data. The inpainting process leverages the feature representations learned by the global model across multiple source domain scenarios, effectively restoring the missing data.

[0120] The patched data is then used to further estimate the power system state. The patched model is transferred to the target domain and fine-tuned using the target domain data. During fine-tuning, only the parameters of a subset of the model layers are adjusted to accommodate the specific data distribution of the target domain.

[0121] The fine-tuned model is used to perform state estimation on the target domain data to ensure that the model can accurately predict the operating state of the power system. State estimation includes the prediction of critical states of the power system, such as abnormal or emergency states.

[0122] In order to intuitively demonstrate the performance of the federated learning-based autoencoder model in power system data patching, the present invention randomly selected a set of masked data from the validation set and compared its predicted values ​​with the true values. Figure 2 The repair effect of P (active power) and Q (reactive power) data is shown. Figure 2 The accuracy and performance of the model when inpainting missing data can be clearly seen. To evaluate the effectiveness of the inpainting, we compared the true values ​​of randomly selected P and Q data with the predicted values ​​after inpainting. The blue dots in the figure represent the true values, and the green dots represent the inpainted values ​​predicted by the model. We compared the masked data points and demonstrated the inpainting effect.

[0123] In the state estimation effect verification, the present invention randomly selected some samples with missing data from the verification set, and compared the estimated values ​​of active power (P), reactive power (Q), voltage (V) and phase angle (θ) with the true values. By drawing the comparison chart, we can intuitively show the estimation effect of the model in the case of missing data, such as Figure 3 Each figure compares the true value of the masked data (blue) and the model estimation value (red), and the results show that the model performs well in estimating P and Q, V and θ.

[0124] The present invention also provides an autoencoder static distribution network state estimation system based on federated learning and transfer learning, comprising:

[0125] The simulation data acquisition module inputs the static distribution network system simulation model for simulation when part of the input data is missing, and collects simulation data of the static distribution network system simulation model;

[0126] A repair model generation module generates a repair model based on simulation data of a static distribution network system simulation model;

[0127] The transfer learning module obtains the target domain model through transfer learning fine-tuning based on the repair model;

[0128] The state estimation module uses the repair model to repair the actual input data with some missing data in the actual static distribution network system to obtain complete input data. Based on the complete input data, the target domain model is used to estimate the operating state of the actual static distribution network system.

[0129] The present invention also provides a computer-readable storage medium on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0130] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0131] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0134] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention shall still fall within the scope of protection of the present invention.

Claims

1. A method for estimating the state of a static distribution network using an autoencoder, characterized in that: include: When part of the input data is missing, the static distribution network system simulation model is input for simulation, the simulation data of the static distribution network system simulation model is collected, and a repair model is generated based on the simulation data of the static distribution network system simulation model; Based on the repair model, the target domain model is obtained through transfer learning fine-tuning; Based on the actual input data with some missing data in the actual static distribution network system, the repair model is used to repair the complete input data. Based on the complete input data, the target domain model is used to estimate the operating status of the actual static distribution network system. The repair model is generated as follows: Within the framework of federated learning, the simulation data of the static distribution network system is clustered using the Gaussian mixture model (GMM). The data is divided into multiple subsets, and the data of each subset is assigned to a different federated learning client. Each federated learning client independently trains an autoencoder model, and the weights of the trained autoencoder model are sent to a central server. The central server aggregates the weights of the autoencoder models of all federated learning clients through weighted averaging to generate a repair model. Based on the repair model, the specific implementation method of obtaining the target domain model through transfer learning fine-tuning is as follows: In transfer learning, let subscripts S and T represent the source task and target task respectively, D S and D T They are the source domain and the target domain respectively, and the corresponding domains are constructed using the feature vector space X and the probability distribution function P(X), namely: Let Y s and Y t are the label vector spaces of the source domain and the target domain respectively, f s and f t are the mapping functions of the source domain and the target domain respectively, then the tasks T in the source domain and the target domain are s and T t They are described as: Assuming that the source domain and the target domain have the same feature distribution but different label spaces, the goal of transfer learning is to learn the mapping function f t :X t →X s In D T The expected error on X is the smallest and satisfies X t =X s 、Y t =Y s and P(Y t |X t )≠P(Y s |X s ); The repair model is used as the source domain model. Through transfer learning, the feature representation and parameters of the repair model are transferred to the target domain model. The parameters are fine-tuned on the target domain model so that the parameters on the target domain model adapt to the characteristics of the target domain data. The target domain model with fine-tuned parameters is trained on the target domain data to obtain the final target domain model.

2. The method for estimating the state of an autoencoder static distribution network according to claim 1, characterized in that: Some data are missing, including setting the active power P and reactive power Q in the input data to zero.

3. The method for estimating the state of an autoencoder static distribution network according to claim 1, characterized in that: The encoder of the autoencoder model consists of four fully connected layers, which gradually compresses the dimensions of the input data to a low-dimensional representation; the decoder of the autoencoder model reconstructs the reconstructed data with the same dimension as the input data by expanding the compressed low-dimensional representation layer by layer.

4. The method for estimating the state of an autoencoder static distribution network according to claim 1 or 3, characterized in that: The autoencoder model is trained as follows: The encoder of the autoencoder model converts the input data x into hidden variables z, which can be expressed as: z=f(w (1) x+b (1) ) Where w (1) represents the weight matrix; b (1) represents the bias vector; f() represents the activation function; The decoder of the autoencoder model reconstructs data that is similar to the input data x based on the hidden variable z It can be expressed as: Where w (2) represents the weight matrix; b (2) represents the bias vector; The optimization goal of the autoencoder model is to ensure that the reconstructed data As close as possible to the input data x; given a set of input data x (n) ∈R (D) ,1≤n≤N, the reconstruction error is expressed as: Where, Represents reconstructed data; By minimizing the reconstruction error, the network parameters of the autoencoder model are learned as θ = {w (1) ,b (1) ,w (2) ,b (2) }.

5. The method for estimating the state of an autoencoder static distribution network according to claim 4, characterized in that: The activation function uses the ReLU function: f(Z)=max(0,Z) Z represents input.

6. The method for estimating the state of an autoencoder static distribution network according to claim 1, characterized in that: The central server aggregates the weights of the autoencoder models of all federated learning clients by weighted averaging as follows: Suppose there are K federated learning clients, whose local data are represented as D1, D2, ..., D k ,…,D K , for the dataset D of the kth federated learning client k , the collected loss function is: Where, f j (w) is f(x j ,y j ), x j represents the input variable of the jth sample; y j represents the output variable of the jth sample; the loss function defined on the dataset set of all federated learning clients is: Where, |D k | indicates D k The number of samples in |D| represents the total number of samples in all federated learning clients. The global aggregation parameter set w(t) of the central server at time t is expressed as follows: Where μ is the learning rate; is the local gradient of the loss function; w k (t) is the local parameter set of the kth federated learning client at time t, expressed as: The central server obtains the local gradient of the loss function of the kth federated learning client at time t Calculate w(t).

7. An autoencoder static distribution network state estimation system using the autoencoder static distribution network state estimation method according to claim 1, characterized in that: include: The simulation data acquisition module inputs the static distribution network system simulation model for simulation when part of the input data is missing, and collects simulation data of the static distribution network system simulation model; A repair model generation module generates a repair model based on simulation data of a static distribution network system simulation model; The transfer learning module obtains the target domain model through transfer learning fine-tuning based on the repair model; The state estimation module uses the repair model to repair the actual input data with some missing data in the actual static distribution network system to obtain complete input data. Based on the complete input data, the target domain model is used to estimate the operating state of the actual static distribution network system.

8. A computer-readable storage medium storing computer program instructions that can be executed by a processor, wherein when the processor executes the computer program instructions, the method steps according to any one of claims 1 to 6 can be implemented.

Images