CNN-Attention-based power distribution network pseudo measurement data prediction method and system

Abstract

The invention discloses a CNN-Attention-based power distribution network pseudo measurement data prediction method and system. The method comprises the steps of constructing a training data set according to pseudo label data and historical measurement data; clustering the training data set, and constructing typical load power curve template library data according to a clustering result; constructing a CNN-Attention-based power distribution network pseudo measurement generation model, and training the power distribution network pseudo measurement generation model according to the typical load power curve template library data; in an actual prediction process, deploying the trained power distribution network pseudo measurement generation model to each node to be measured in a target power distribution network, and adjusting the trained power distribution network pseudo measurement generation model through a transfer learning algorithm; and inputting the obtained real-time measurement data of each measured node into the adjusted pseudo-measurement generation model of the power distribution network, and predicting to obtain pseudo-measurement data of each node to be measured. According to the method provided by the invention, the technical problem of limited data acquisition density of the power distribution network is solved.

Description

Technical Field

[0001] The present invention relates to the field of smart grid and load management of power systems, and in particular to a distribution network pseudo-measurement data prediction method and system based on CNN-Attention. Background Art

[0002] As the level of intelligence and automation of power systems continues to increase, the safe and stable operation of distribution networks has become increasingly important for accurate power information and data analysis. By deploying a large number of sensors and smart devices, smart distribution networks can monitor the operating status of the power grid in real time. Combined with advanced data analysis technology, the system can promptly detect potential fault hazards and issue early warnings, thereby effectively avoiding large-scale power outages.

[0003] Traditional distribution network sensing systems usually rely on a limited number of actual sensors to obtain power data. The layout of these sensors is often limited by the difficulty of installation, including terrain and environmental factors, infrastructure limitations, resource limitations, and technical and standard limitations. These factors work together to result in limited data collection density in the distribution network. Summary of the invention

[0004] The present invention aims to provide a distribution network pseudo-measurement data prediction method and system based on CNN-Attention to solve the technical problem of limited data acquisition density in the distribution network, thereby achieving the effect of promoting further improvement of the intelligence and automation level of the distribution network.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a distribution network pseudo-measurement data prediction method based on CNN-Attention, comprising:

[0006] Obtain historical measurement data of measured nodes with sensors installed in the target distribution network;

[0007] The pre-built LSTM model is trained by the historical measurement data to obtain an initial LSTM model, each node to be measured in the target distribution network is predicted by the initial LSTM model to obtain pseudo-label data, and a training data set is constructed according to the pseudo-label data and the historical measurement data; the node to be measured is designed to be related to the power of the measured node;

[0008] Clustering the training data set, and constructing typical load power curve template library data according to the clustering results;

[0009] Constructing a distribution network pseudo-measurement generation model based on CNN-Attention, training the distribution network pseudo-measurement generation model according to the typical load power curve template library data, and obtaining the trained distribution network pseudo-measurement generation model;

[0010] In the actual prediction process, the trained distribution network pseudo-measurement generation model is deployed to each of the nodes to be measured in the target distribution network, and the trained distribution network pseudo-measurement generation model is adjusted through a transfer learning algorithm to make the distribution network pseudo-measurement generation model adapt to the data distribution of the nodes to be measured;

[0011] The acquired real-time measurement data of each of the measured nodes is input into the adjusted distribution network pseudo-measurement generation model to predict and obtain the pseudo-measurement data of each of the nodes to be measured.

[0012] As one of the preferred solutions, clustering the training data set by a clustering algorithm and constructing a typical load power curve template library data according to the clustering result includes:

[0013] Initializing the training data set and randomly selecting a number of data points as initial cluster centers of the training data set;

[0014] During the allocation process, the distance from each data point in the training data set to each initial cluster center is calculated respectively, and the corresponding initial cluster center is allocated to each data point according to the distance result to obtain a plurality of initial clusters;

[0015] During the calculation process, the cluster center of each of the initial clusters is calculated respectively;

[0016] Repeating the execution allocation process and the calculation process until a preset number of iterations is reached;

[0017] The cluster centers after iteration are processed, and typical load power curve template library data is constructed according to the processing results.

[0018] As one of the preferred solutions, the distribution network pseudo measurement generation model is trained according to the typical load power curve template library to obtain the trained distribution network pseudo measurement generation model, including:

[0019] Randomly initializing the parameters of the distribution network pseudo-measurement generation model;

[0020] In the model calculation process, the output and actual loss of the distribution network pseudo-measurement generation model are calculated according to the typical load power curve template library, and the loss function of the distribution network pseudo-measurement generation model is calculated according to the output and the actual loss;

[0021] In the parameter updating process, the parameters are updated according to the calculation result of the loss function and the gradient descent method;

[0022] The execution model calculation process and the parameter updating process are repeated until a preset number of iterations is reached to obtain the trained distribution network pseudo-measurement generation model.

[0023] As one of the preferred solutions, the loss function is expressed as:

[0024]

[0025] Among them, MSE(θ) is the loss function, N is the number of sampling points, y i Actual active power measurement, is the generated pseudo measurement.

[0026] As one preferred solution, the historical measurement data includes: active power, voltage amplitude and current amplitude;

[0027] After obtaining the historical measurement data of the measured nodes of the target distribution network where the sensors are set, the method further includes preprocessing the historical measurement data:

[0028] Performing time alignment on the historical measurement data of different acquisition frequencies;

[0029] The historical measurement data after time alignment is subjected to data normalization processing.

[0030] As one preferred solution, after constructing a training data set according to the pseudo-label data and the pre-processed historical measurement data, the method further includes:

[0031] Iteratively training the initial LSTM model using the training data set;

[0032] The accuracy of the LSTM model obtained through iterative training is evaluated through a validation set, and the LSTM model is optimized according to the evaluation result.

[0033] As one of the preferred solutions, the method of adjusting the trained distribution network pseudo-measurement generation model by using a transfer learning algorithm includes:

[0034] Evaluate the difference between the data to be measured and the historical measurement data based on the maximum mean difference (MMD);

[0035] The parameters of the distribution network pseudo-measurement generation model are frozen and adjusted according to the difference.

[0036] The freezing and adjusting the parameters of the distribution network pseudo-measurement generation model based on the difference includes:

[0037] As one of the preferred solutions, freezing and adjusting the parameters of the distribution network pseudo-measurement generation model based on the difference includes:

[0038] Based on the difference and the structure of the distribution network pseudo-measurement generation model, selecting a frozen layer of the distribution network pseudo-measurement generation model;

[0039] Freezing parameters of the distribution network pseudo-measurement generation model according to the selected freezing layer;

[0040] The frozen distribution network pseudo-measurement generation model is trained based on part of the node data to be measured that is prepared in advance, and the distribution network pseudo-measurement generation model is adjusted based on the parameters obtained during the training process.

[0041] As one preferred solution, after adjusting the distribution network pseudo-measurement generation model based on the parameters obtained in the training process, the method further includes:

[0042] Evaluating the performance of the adjusted distribution network pseudo-measurement generation model through a test set;

[0043] The model is iteratively optimized according to the evaluation results, and the optimization method includes adjusting the model structure, parameters and training strategy.

[0044] Another embodiment of the present invention provides a distribution network pseudo-measurement data prediction system based on CNN-Attention, including:

[0045] An acquisition module, used to acquire historical measurement data of measured nodes of which sensors are set in the target distribution network;

[0046] A construction module is used to train a pre-constructed LSTM model through the historical measurement data to obtain an initial LSTM model, predict each node to be measured in the target distribution network through the initial LSTM model to obtain pseudo-label data, and construct a training data set according to the pseudo-label data and the historical measurement data; the node to be measured is designed to be related to the active power of the measured node;

[0047] A clustering module, used for clustering the training data set and building a typical load power curve template library data according to the clustering results;

[0048] A training module, used for constructing a distribution network pseudo-measurement generation model based on CNN-Attention, training the distribution network pseudo-measurement generation model according to the typical load power curve template library data, and obtaining the trained distribution network pseudo-measurement generation model;

[0049] A migration module is used to deploy the trained distribution network pseudo-measurement generation model to each of the nodes to be measured in the target distribution network during the actual prediction process, and adjust the trained distribution network pseudo-measurement generation model through a migration learning algorithm so that the distribution network pseudo-measurement generation model adapts to the data distribution of the nodes to be measured;

[0050] The prediction module is used to input the acquired real-time measurement data of each node into the adjusted distribution network pseudo-measurement generation model to predict and obtain the pseudo-measurement data of each node to be measured.

[0051] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0052] (1) By generating virtual sensor data, this technology significantly improves the density and accuracy of distribution network data collection, thereby enhancing the reliability of power information.

[0053] (2) Combined with the transfer learning method, the system can efficiently and adaptively perform model optimization methods and parameter adjustment strategies to generate pseudo-measurement data for each node in the distribution network, effectively covering areas that are difficult for sensors to reach using traditional methods;

[0054] (3) The use of advanced CNN_Attention mechanism has made important contributions to improving the intelligence and automation level of distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of a method for predicting pseudo-measurement data of a distribution network based on CNN-Attention in one embodiment of the present invention;

[0056] Figure 2 It is a network structure diagram of a distribution network pseudo-measurement generation model based on CNN-Attention in one embodiment of the present invention;

[0057] Figure 3 It is a schematic diagram of a distribution network pseudo-measurement data prediction system based on CNN-Attention in one embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0060] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0061] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0062] An embodiment of the present invention provides a distribution network pseudo-measurement data prediction method based on CNN-Attention. For details, see Figure 1 , Figure 1 The flowchart of the pseudo-measurement data prediction method for distribution network based on CNN-Attention provided in one embodiment of the present invention is shown. It includes steps S1 to S6:

[0063] S1. Obtain historical measurement data of measured nodes with sensors installed in the target distribution network;

[0064] Collect historical measurement data of each branch and node of the distribution network through sensors, including active power, voltage amplitude, current amplitude, etc.;

[0065] Preferably, in an embodiment of the present invention, the historical measurement data includes: active power, voltage amplitude and current amplitude;

[0066] After obtaining the historical measurement data of the measured nodes of the target distribution network where the sensors are set, the method further includes preprocessing the historical measurement data:

[0067] Performing time alignment on the historical measurement data of different acquisition frequencies;

[0068] The historical measurement data after time alignment is subjected to data normalization processing.

[0069] The purpose of time alignment is to ensure that all data sets have corresponding measurement values ​​at the same time point or time period so that they can be compared and analyzed; data normalization is to convert all data into the same numerical range (usually between 0 and 1 or between -1 and 1) so that the effects of different dimensions and magnitudes can be eliminated during comparison and analysis.

[0070] Specifically, for the i-th historical measurement value at the k-th time step, the historical measurement value is normalized. The specific formula is as follows:

[0071]

[0072] Among them, x k ′ is the kth normalized historical measurement value, x k is the kth original historical measurement value, x mink is the minimum value among the k groups of data, x max k is the maximum value of a set of data.

[0073] S2. The pre-built LSTM model is trained by the historical measurement data to obtain an initial LSTM model, and each node to be measured in the target distribution network is predicted by the initial LSTM model to obtain pseudo-label data, and a training data set is constructed according to the pseudo-label data and the historical measurement data; the node to be measured is designed to be related to the power of the measured node;

[0074] Generating virtual sensor data to enhance the density and accuracy of distribution network data collection is an effective data enhancement strategy. Combining LSTM for prediction and using the prediction results as pseudo labels to construct training data sets can further improve the model's prediction ability and robustness.

[0075] Retrain the LSTM model using a dataset with pseudo labels. Since pseudo labels are generated based on model predictions, this process can be considered a form of self-supervised learning or weakly supervised learning. By iterating this process, the prediction ability of the LSTM model can be gradually improved.

[0076] Preferably, in an embodiment of the present invention, after constructing a training data set according to the pseudo-label data and the pre-processed historical measurement data, it further includes:

[0077] Iteratively training the initial LSTM model using the training data set;

[0078] The accuracy of the LSTM model obtained through iterative training is evaluated through a validation set, and the LSTM model is optimized according to the evaluation result.

[0079] S3, clustering the training data set, and constructing typical load power curve template library data according to the clustering results;

[0080] A small amount of measured data is used as clustering centers to cluster the labeled data to build a typical load power curve template library.

[0081] Preferably, in an embodiment of the present invention, clustering the training data set by a clustering algorithm and constructing a typical load power curve template library data according to the clustering result includes:

[0082] Initializing the training data set and randomly selecting a number of data points as initial cluster centers of the training data set;

[0083] During the allocation process, the distance from each data point in the training data set to each initial cluster center is calculated respectively, and the corresponding initial cluster center is allocated to each data point according to the distance result to obtain a plurality of initial clusters;

[0084] During the calculation process, the cluster center of each of the initial clusters is calculated respectively;

[0085] Repeating the execution allocation process and the calculation process until a preset number of iterations is reached;

[0086] The cluster centers after iteration are processed, and typical load power curve template library data is constructed according to the processing results.

[0087] Specifically, the clustering algorithm groups similar load power curves into the same cluster. For each cluster, one or more representative load power curves are selected as templates for the cluster. This can be achieved by calculating the mean, median, or selecting curves near the center of the cluster. The representative curves of all clusters are combined to construct a typical load power curve template library. This template library will contain a series of representative load power curves that can cover the main load patterns in the training data set.

[0088] S4. Constructing a distribution network pseudo-measurement generation model based on CNN-Attention, training the distribution network pseudo-measurement generation model according to the typical load power curve template library data, and obtaining the trained distribution network pseudo-measurement generation model;

[0089] See also Figure 2 , Figure 2 Shown is a network structure diagram of a distribution network pseudo-measurement generation model based on CNN-Attention provided in one embodiment of the present invention.

[0090] First, the model parameters are randomly initialized. This includes selecting an initialization strategy, setting weights, learning rates, and selecting an optimizer. Select appropriate initialization schemes and weight initialization methods for each layer's activation function and network characteristics. In this embodiment, the Xavier method is used for random initialization, the learning rate is set to 0.001, and the optimizer uses Adam.

[0091] Preferably, in an embodiment of the present invention, the training of the distribution network pseudo-measurement generation model according to the typical load power curve template library to obtain the trained distribution network pseudo-measurement generation model includes:

[0092] Randomly initializing the parameters of the distribution network pseudo-measurement generation model;

[0093] In the model calculation process, the output and actual loss of the distribution network pseudo-measurement generation model are calculated according to the typical load power curve template library, and the loss function of the distribution network pseudo-measurement generation model is calculated according to the output and the actual loss;

[0094] In the parameter updating process, the parameters are updated according to the calculation result of the loss function and the gradient descent method;

[0095] The execution model calculation process and the parameter updating process are repeated until a preset number of iterations is reached to obtain the trained distribution network pseudo-measurement generation model.

[0096] Preferably, in an embodiment of the present invention, the loss function is expressed as:

[0097]

[0098] Among them, MSE(θ) is the loss function, N is the number of sampling points, y i Actual active power measurement, is the generated pseudo measurement.

[0099] After calculating L(θ), the parameters are updated by gradient descent.

[0100]

[0101] Wherein, ɑ is the update rate of the target value function neural network. In this embodiment, it is set to 0.005.

[0102] Finally, the test data set is used to evaluate the performance of the model and optimize it to ensure that the model has good performance. In this embodiment, the mean absolute percentage error (MPAE) is used as a performance evaluation method.

[0103] S5. In the actual prediction process, the trained distribution network pseudo-measurement generation model is deployed to each of the nodes to be measured in the target distribution network, and the trained distribution network pseudo-measurement generation model is adjusted through a transfer learning algorithm to make the distribution network pseudo-measurement generation model adapt to the data distribution of the nodes to be measured;

[0104] First, we need to extract high-dimensional features from the source domain (the nodes that have been measured) and the target domain (the nodes to be measured). Secondly, we need to measure the difference between the probability distributions of the two datasets.

[0105] Preferably, in an embodiment of the present invention, the adjusting the trained distribution network pseudo-measurement generation model by using a transfer learning algorithm includes:

[0106] Evaluate the difference between the data to be measured and the historical measurement data based on the maximum mean difference (MMD);

[0107] The parameters of the distribution network pseudo-measurement generation model are frozen and adjusted according to the difference.

[0108] Specifically, MMD evaluates the similarity or dissimilarity between two data sets by comparing their mean differences in a high-dimensional space. The calculation of the MMD value usually involves the following steps:

[0109] Map data samples to high-dimensional space (through kernel function);

[0110] Compute the means of two distributions in a high-dimensional space;

[0111] Calculate the difference between these two means, which is the MMD value.

[0112] After evaluating the difference between the data to be measured and the historical measurement data based on the maximum mean difference (MMD), freezing and adjusting the parameters of the distribution network pseudo-measurement generation model is a key optimization step.

[0113] Preferably, in an embodiment of the present invention, freezing and adjusting the parameters of the distribution network pseudo-measurement generation model based on the difference includes:

[0114] Based on the difference and the structure of the distribution network pseudo-measurement generation model, selecting a frozen layer of the distribution network pseudo-measurement generation model;

[0115] Freezing parameters of the distribution network pseudo-measurement generation model according to the selected freezing layer;

[0116] The frozen distribution network pseudo-measurement generation model is trained based on part of the node data to be measured that is prepared in advance, and the distribution network pseudo-measurement generation model is adjusted based on the parameters obtained during the training process.

[0117] Among them, the parameter freezing strategy includes: if some parameters in the distribution network pseudo-measurement generation model are very sensitive to small changes in the input data, and these changes do not significantly improve the performance of the model, these parameters can be frozen; or when the data difference (MMD) is small, the model may be close to the optimal solution, and excessive adjustment of parameters may destroy the stability of the model. By freezing some parameters, the stability and generalization ability of the model can be maintained.

[0118] Preferably, in an embodiment of the present invention, after adjusting the distribution network pseudo-measurement generation model based on the parameters obtained in the training process, the method further includes:

[0119] Evaluating the performance of the adjusted distribution network pseudo-measurement generation model through a test set;

[0120] The model is iteratively optimized according to the evaluation results, and the optimization method includes adjusting the model structure, parameters and training strategy.

[0121] S6. Input the acquired real-time measurement data of each of the measured nodes into the adjusted distribution network pseudo-measurement generation model to predict and obtain the pseudo-measurement data of each of the nodes to be measured.

[0122] The predicted results are compared with the actual values ​​to evaluate the prediction accuracy of the model, and feedback and adjustments are made to the model based on the results of the performance evaluation.

[0123] Through the above steps, the real-time measurement data of each measured node is input into the adjusted distribution network pseudo measurement generation model, and the pseudo measurement data of each node to be measured is predicted. These data can provide important reference for the monitoring, control and optimization of the distribution network.

[0124] Another embodiment of the present invention provides a distribution network pseudo-measurement data prediction system based on CNN-Attention, see Figure 3 , Figure 3The schematic diagram of a distribution network pseudo-measurement data prediction system based on CNN-Attention provided in one embodiment of the present invention includes: an acquisition module 11, a construction module 12, a clustering module 13, a training module 14, a migration module 15, and a prediction module 16, wherein:

[0125] An acquisition module 11 is used to acquire historical measurement data of measured nodes of which sensors are set in the target distribution network;

[0126] A construction module 12 is used to train a pre-constructed LSTM model through the historical measurement data to obtain an initial LSTM model, predict each node to be measured in the target distribution network through the initial LSTM model to obtain pseudo-label data, and construct a training data set according to the pseudo-label data and the historical measurement data; the node to be measured is designed to be related to the active power of the measured node;

[0127] A clustering module 13 is used to cluster the training data set and construct a typical load power curve template library data according to the clustering results;

[0128] A training module 14 is used to construct a distribution network pseudo-measurement generation model based on CNN-Attention, and train the distribution network pseudo-measurement generation model according to the typical load power curve template library data to obtain the trained distribution network pseudo-measurement generation model;

[0129] The migration module 15 is used to deploy the trained distribution network pseudo-measurement generation model to each of the nodes to be measured in the target distribution network during the actual prediction process, and adjust the trained distribution network pseudo-measurement generation model through a migration learning algorithm so that the distribution network pseudo-measurement generation model adapts to the data distribution of the nodes to be measured;

[0130] The prediction module 16 is used to input the acquired real-time measurement data of each node into the adjusted distribution network pseudo-measurement generation model to predict and obtain the pseudo-measurement data of each node to be measured.

[0131] The distribution network pseudo-measurement data prediction method and system based on CNN_Attention of the present invention have significant beneficial effects. First, by generating virtual sensor data, the technology significantly improves the density and accuracy of distribution network data acquisition, thereby enhancing the reliability of power information. Secondly, combined with the transfer learning method, the system can efficiently and adaptively perform model optimization methods and parameter adjustment strategies to generate pseudo-measurement data for each node in the distribution network, effectively covering areas that are difficult for sensors to reach in traditional methods. In addition, this method and system utilize the advanced CNN_Attention mechanism, which has made important contributions to improving the intelligence and automation level of the distribution network. In summary, the present invention not only optimizes the efficiency and accuracy of deep neural networks, but also enhances the adaptability and application scope of the system, providing an efficient solution for intelligent monitoring and energy management of power systems.

[0132] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A distribution network pseudo-measurement data prediction method based on CNN-Attention, It is characterized in that The following steps are involved: Obtain historical measurement data of measured nodes with sensors installed in the target distribution network; The pre-built LSTM model is trained by the historical measurement data to obtain an initial LSTM model, each node to be measured in the target distribution network is predicted by the initial LSTM model to obtain pseudo-label data, and a training data set is constructed according to the pseudo-label data and the historical measurement data; the node to be measured is designed to be related to the power of the measured node; Clustering the training data set, and constructing typical load power curve template library data according to the clustering results; Constructing a distribution network pseudo-measurement generation model based on CNN-Attention, training the distribution network pseudo-measurement generation model according to the typical load power curve template library data, and obtaining the trained distribution network pseudo-measurement generation model; In the actual prediction process, the trained distribution network pseudo-measurement generation model is deployed to each of the nodes to be measured in the target distribution network, and the trained distribution network pseudo-measurement generation model is adjusted through a transfer learning algorithm to make the distribution network pseudo-measurement generation model adapt to the data distribution of the nodes to be measured; The acquired real-time measurement data of each of the measured nodes is input into the adjusted distribution network pseudo-measurement generation model to predict and obtain the pseudo-measurement data of each of the nodes to be measured.

2. According to claim 1, the distribution network pseudo-measurement data prediction method based on CNN-Attention, It is characterized in that The clustering of the training data set by a clustering algorithm and the construction of a typical load power curve template library data according to the clustering result include: Initializing the training data set and randomly selecting a number of data points as initial cluster centers of the training data set; During the allocation process, the distance from each data point in the training data set to each initial cluster center is calculated respectively, and the corresponding initial cluster center is allocated to each data point according to the distance result to obtain a plurality of initial clusters; During the calculation process, the cluster center of each of the initial clusters is calculated respectively; Repeating the execution allocation process and the calculation process until a preset number of iterations is reached; The cluster centers after iteration are processed, and typical load power curve template library data is constructed according to the processing results.

3. According to claim 1, the distribution network pseudo-measurement data prediction method based on CNN-Attention, It is characterized in that The training of the distribution network pseudo measurement generation model according to the typical load power curve template library to obtain the trained distribution network pseudo measurement generation model includes: Randomly initializing the parameters of the distribution network pseudo-measurement generation model; In the model calculation process, the output and actual loss of the distribution network pseudo-measurement generation model are calculated according to the typical load power curve template library, and the loss function of the distribution network pseudo-measurement generation model is calculated according to the output and the actual loss; In the parameter updating process, the parameters are updated according to the calculation result of the loss function and the gradient descent method; The execution model calculation process and the parameter updating process are repeated until a preset number of iterations is reached to obtain the trained distribution network pseudo-measurement generation model.

4. According to claim 3, the distribution network pseudo-measurement data prediction method based on CNN-Attention, It is characterized in that The loss function is expressed as: Among them, MSE(θ) is the loss function, N is the number of sampling points, y i Actual active power measurement, is the generated pseudo measurement.

5. According to claim 1, the distribution network pseudo-measurement data prediction method based on CNN-Attention, It is characterized in that The historical measurement data include: active power, voltage amplitude and current amplitude; After acquiring the historical measurement data of the measured nodes of the target distribution network where the sensors are set, the method further includes preprocessing the historical measurement data, wherein the preprocessing of the historical measurement data includes: Performing time alignment on the historical measurement data of different acquisition frequencies; The historical measurement data after time alignment is subjected to data normalization processing.

6. According to claim 1, the method for predicting pseudo-measurement data of distribution network based on CNN-Attention, It is characterized in that After constructing a training data set according to the pseudo-label data and the pre-processed historical measurement data, the method further includes: Iteratively training the initial LSTM model using the training data set; The accuracy of the LSTM model obtained through iterative training is evaluated through a validation set, and the LSTM model is optimized according to the evaluation result.

7. The distribution network pseudo-measurement data prediction method based on CNN-Attention according to claim 1, It is characterized in that The step of adjusting the trained distribution network pseudo-measurement generation model by using a transfer learning algorithm includes: Evaluate the difference between the data to be measured and the historical measurement data based on the maximum mean difference (MMD); The parameters of the distribution network pseudo-measurement generation model are frozen and adjusted according to the difference.

8. According to claim 7, the distribution network pseudo-measurement data prediction method based on CNN-Attention, It is characterized in that The freezing and adjusting the parameters of the distribution network pseudo-measurement generation model according to the difference includes: Based on the difference and the structure of the distribution network pseudo-measurement generation model, selecting a frozen layer of the distribution network pseudo-measurement generation model; Freezing parameters of the distribution network pseudo-measurement generation model according to the selected freezing layer; The frozen distribution network pseudo-measurement generation model is trained based on part of the node data to be measured that is prepared in advance, and the distribution network pseudo-measurement generation model is adjusted based on the parameters obtained during the training process.

9. The distribution network pseudo-measurement data prediction method based on CNN-Attention according to claim 8, It is characterized in that After adjusting the distribution network pseudo-measurement generation model based on the parameters obtained in the training process, the method further includes: Evaluating the performance of the adjusted distribution network pseudo-measurement generation model through a test set; The model is iteratively optimized according to the evaluation results, and the optimization method includes adjusting the model structure, parameters and training strategy.

10. A distribution network pseudo-measurement data prediction system based on CNN-Attention, It is characterized in that Includes the following modules: An acquisition module, used to acquire historical measurement data of measured nodes of which sensors are set in the target distribution network; A construction module is used to train a pre-constructed LSTM model through the historical measurement data to obtain an initial LSTM model, predict each node to be measured in the target distribution network through the initial LSTM model to obtain pseudo-label data, and construct a training data set according to the pseudo-label data and the historical measurement data; the node to be measured is designed to be related to the active power of the measured node; A clustering module, used for clustering the training data set and building a typical load power curve template library data according to the clustering results; A training module, used for constructing a distribution network pseudo-measurement generation model based on CNN-Attention, training the distribution network pseudo-measurement generation model according to the typical load power curve template library data, and obtaining the trained distribution network pseudo-measurement generation model; A migration module is used to deploy the trained distribution network pseudo-measurement generation model to each of the nodes to be measured in the target distribution network during the actual prediction process, and adjust the trained distribution network pseudo-measurement generation model through a migration learning algorithm so that the distribution network pseudo-measurement generation model adapts to the data distribution of the nodes to be measured; The prediction module is used to input the acquired real-time measurement data of each node into the adjusted distribution network pseudo-measurement generation model to predict and obtain the pseudo-measurement data of each node to be measured.

Images