A distribution network voltage transformer state evaluation method and device

Abstract

The present invention relates to the technical field of online monitoring of electric power metering, and specifically to a method and device for evaluating the state of a distribution network voltage transformer, the method comprising the following steps: segmenting endogenous data and then embedding variables, using a joint self-attention mechanism and fusing it with itself to input a linear network layer to obtain endogenous intermediate features, using a linear network layer to embed variables in exogenous data to obtain exogenous intermediate features respectively; applying a cross-attention mechanism to process the endogenous and exogenous intermediate features, and inputting the results into an MLP network to obtain a predicted value of the actual transformer ratio; performing multi-task learning on the predicted value of the actual transformer ratio to obtain the actual transformer ratio; calculating the current actual error of the transformer according to the actual transformer ratio combined with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal. It can solve the problem that power-off monitoring is required in the prior art, which will lead to invalid maintenance or increase the power outage time.

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring of electric power metering, and in particular to a method and device for evaluating the state of a distribution network voltage transformer. Background Art

[0002] The voltage transformer in the distribution network is used to isolate high voltage and convert it into low voltage, so that relay protection, automatic equipment and metering instruments can obtain the primary voltage signal. The distribution network voltage transformer can monitor the voltage in the distribution network and trigger the protection device when the voltage is too high or too low to prevent equipment damage or failure; at the same time, it can accurately measure the voltage in the distribution network to calculate the consumed electric energy. By monitoring the voltage changes in the distribution network, the voltage transformer can also help identify and locate faults in the power grid, thereby speeding up fault recovery and reducing power outage time.

[0003] However, in the actual operation process of the voltage transformer in the distribution network, its error is affected by the acquisition principle and harsh environment, which may cause failure within its working life. Therefore, it is necessary to evaluate the operating status of the voltage transformer so that it can be quickly identified when an abnormality occurs, which will help relevant operation and maintenance personnel to carry out inspection and maintenance to ensure the stable operation of the power grid.

[0004] However, in the prior art, inaccurate status assessment of the voltage transformer or power failure monitoring may cause operation and maintenance personnel to perform ineffective inspections and maintenance or increase power outage time. Summary of the invention

[0005] In view of the defects existing in the prior art, the purpose of the present invention is to provide a distribution network voltage transformer status assessment method and device, which can solve the problem of inaccurate status assessment or power-off monitoring of voltage transformers in the prior art, which will cause operation and maintenance personnel to perform invalid inspections and maintenance or increase power outage time.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0007] In one aspect, the present invention provides a method for evaluating the state of a voltage transformer in a distribution network, comprising the following steps:

[0008] The endogenous data is segmented and then embedded into variables, and the joint self-attention mechanism is used to fuse it with itself and input into the linear network layer to obtain the endogenous intermediate features. The exogenous data is embedded into the linear network layer to obtain the exogenous intermediate features. The endogenous data includes the transformer low-voltage side voltage data, and the exogenous data includes the low-voltage side load data, temperature data and humidity data.

[0009] The cross-attention mechanism is applied to the endogenous intermediate features and the exogenous intermediate features to obtain the endogenous and exogenous correlation features, which are input into the MLP network for processing to obtain the output features as the actual transformer ratio prediction value;

[0010] Multi-task learning of regression task and classification task is performed on the predicted value of the actual transformer ratio to obtain the actual transformer ratio;

[0011] The voltage on the high-voltage side of the transformer is calculated based on the actual transformation ratio of the transformer, and the current actual error of the transformer is calculated in combination with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal.

[0012] In some optional schemes, the endogenous data is segmented and then embedded into variables, and the endogenous intermediate features are obtained by using the joint self-attention mechanism and fusing it with itself into a linear network layer, and the exogenous data is embedded into variables using a linear network layer to obtain exogenous intermediate features, including:

[0013] The transformer low-voltage side voltage data is evenly divided to obtain low-voltage side voltage division data;

[0014] The low-voltage side voltage segmentation data is subjected to the joint self-attention mechanism and fused with its own input linear network layer to obtain the intermediate features of the low-voltage side voltage;

[0015] The low-voltage side load data, temperature data and humidity data are embedded in the linear network layer to obtain the intermediate features of the low-voltage side load, the intermediate features of the temperature and the intermediate features of the humidity respectively.

[0016] In some optional schemes, the cross-attention mechanism is applied to the endogenous intermediate features and the exogenous intermediate features to obtain the endogenous and exogenous correlation features, and the features are input into the MLP network for processing to obtain the output features as the actual transformer ratio prediction value, including:

[0017] The cross-attention mechanism is applied to the intermediate voltage features of the low-voltage side and the intermediate load features, temperature intermediate features and humidity intermediate features of the low-voltage side to obtain load voltage correlation features, temperature voltage correlation features and humidity voltage correlation features;

[0018] The load voltage correlation characteristics, temperature voltage correlation characteristics and humidity voltage correlation characteristics are input into the MLP network for processing, and the output characteristics are obtained as the prediction value of the actual transformation ratio of the transformer.

[0019] In some optional schemes, the multi-task learning of the regression task and the classification task is performed on the predicted value of the actual transformation ratio of the transformer at the same time to obtain the actual transformation ratio of the transformer, including:

[0020] Use the actual transformer ratio prediction value The activation function obtains the predicted probability value, and the predicted probability value uses the cross entropy loss function for classification task learning;

[0021] The predicted value of the transformer actual transformation ratio is used for regression task learning using the mean square error loss function;

[0022] The total model loss is obtained by weighted summing the cross entropy loss of the classification task and the mean square error loss of the regression task;

[0023] The model parameters are adjusted iteratively in the direction of reducing the total loss of the model to obtain the actual transformation ratio of the transformer.

[0024] In some optional schemes, a dynamic weight averaging algorithm is used to adjust the weight coefficients of the regression task and the classification task.

[0025] In some alternative schemes, according to and Adjust the weight coefficients of regression and classification tasks , Indicates the number of single tasks. In the current model ; is the number of iterations of the model; Indicates the task exist The loss value at the iteration; Indicates the task exist The training speed at iterations; To set a constant.

[0026] In some alternative schemes, according to the formula Calculate the voltage on the high voltage side of the transformer ,in, is the voltage on the high voltage side of the transformer, is the actual transformation ratio of the transformer.

[0027] In some alternative schemes, according to the formula Calculate the rated voltage of the high voltage side of the transformer ,in, is the voltage data of the low voltage side, is the rated transformation ratio.

[0028] In some alternative schemes, according to the formula , calculate the actual current error of the transformer.

[0029] On the other hand, the present invention also provides a distribution network voltage transformer state assessment device, comprising:

[0030] The data processing module is used to segment the endogenous data and then embed the variables, and use the joint self-attention mechanism to fuse the data with itself and input the linear network layer to obtain the endogenous intermediate features, and use the linear network layer to embed the variables of the exogenous data to obtain the exogenous intermediate features, wherein the endogenous data includes the transformer low-voltage side voltage data, and the exogenous data includes the low-voltage side load data, temperature data and humidity data;

[0031] The prediction value output module is used to process the endogenous intermediate features and the exogenous intermediate features by applying a cross-attention mechanism to obtain the endogenous and exogenous correlation features, and input them into the MLP network for processing to obtain the output features as the actual transformer ratio prediction value;

[0032] A learning module, which is used to simultaneously perform multi-task learning of regression tasks and classification tasks on the predicted value of the actual transformation ratio of the transformer to obtain the actual transformation ratio of the transformer;

[0033] The judgment module is used to calculate the voltage on the high-voltage side of the transformer according to the actual transformation ratio of the transformer, and calculate the current actual error of the transformer in combination with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal.

[0034] Compared with the prior art, the advantages of the present invention are: this solution divides the transformer data into endogenous data and exogenous data, extracts the internal dependency of the endogenous data, and calculates the coupling relationship between the exogenous data and the endogenous data as a whole, and the endogenous data is combined with the exogenous data to improve the accuracy of the model. By segmenting the endogenous data and applying a joint self-attention mechanism to learn the internal association of the endogenous data, the amount of calculation is reduced and the calculation efficiency of the model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 Schematic diagram of the flow of a method for evaluating the state of a voltage transformer in a distribution network according to an embodiment of the present invention;

[0037] Figure 2 A diagram showing the concept of status evaluation of a voltage transformer in a distribution network according to an embodiment of the present invention;

[0038] Figure 3 A topological relationship diagram of a distribution network transformer and a mutual inductor in an embodiment of the present invention;

[0039] Figure 4Schematic diagram of the joint self-attention mechanism in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0041] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings.

[0042] like Figure 1 , Figure 2 and Figure 3 As shown, on the one hand, the present invention provides a method for evaluating the state of a distribution network voltage transformer, comprising the following steps:

[0043] S0: Collect the measured transformation ratio data, low-voltage side voltage data, low-voltage side load data, temperature data and humidity data of the distribution network transformer within a certain period of time.

[0044] Specifically, collect the measured transformation ratio of the distribution network transformer within a certain period of time , low voltage side voltage data , low voltage side load data , Temperature data and humidity data .

[0045] S1: The endogenous data is split and then embedded into variables. The joint self-attention mechanism is used to fuse the data with itself and input into the linear network layer to obtain the endogenous intermediate features. The exogenous data is embedded into the linear network layer to obtain the exogenous intermediate features. The endogenous data includes the transformer low-voltage side voltage data, and the exogenous data includes the low-voltage side load data, temperature data and humidity data.

[0046] In some optional embodiments, step S1 includes:

[0047] S11: Voltage data of transformer low voltage side , evenly divide to obtain the low voltage side voltage division data .

[0048] In this example, the transformer internal source data, that is, the transformer low-voltage side voltage data Evenly divided into k segments, we get , It is the kth segment after the transformer low-voltage side voltage data is evenly segmented.

[0049] like Figure 4 As shown, S12: The low-voltage side voltage segmentation data is subjected to a joint self-attention mechanism and is fused with its own input linear network layer to obtain the intermediate features of the low-voltage side voltage.

[0050] In this example, the low voltage side voltage segmentation data Use linear network layers to embed variables , for Using linear layers to embed variables The corresponding data.

[0051] right Apply joint self-attention mechanism to obtain transition features :

[0052]

[0053]

[0054]

[0055] in, is the activation function, represents transpose, for The number of queries, , They are The keys and values ​​of for Dimensions; for The number of queries, , They are The keys and values ​​of for Dimensions; . With features After fusion, the linear layer is input to obtain the intermediate characteristics of the low-voltage side voltage , is the self-attention mechanism function, atten_k-1(...,atten_2(atten_1(,)),)...,) is the joint self-attention mechanism function.

[0056] S13: Use a linear network layer to perform variable embedding on the low-voltage side load data, temperature data and humidity data to obtain the low-voltage side load intermediate features, temperature intermediate features and humidity intermediate features respectively.

[0057] In this example, the external data includes the low-voltage side load data , Temperature data and humidity data Use linear network layer to embed variables and obtain the intermediate features of low-voltage side load , Temperature intermediate characteristics , humidity intermediate characteristics .

[0058] S2: The cross-attention mechanism is applied to the endogenous intermediate features and the exogenous intermediate features to obtain the endogenous and exogenous correlation features, which are input into the MLP network for processing to obtain the output features and serve as the actual transformation ratio prediction value of the transformer.

[0059] In some optional embodiments, step S2 includes:

[0060] S21: Apply the cross attention mechanism to the intermediate features of the low-voltage side voltage, the intermediate features of the low-voltage side load, the intermediate features of temperature, and the intermediate features of humidity to obtain the load-voltage correlation features. , Temperature-voltage correlation characteristics Humidity-voltage correlation characteristics .

[0061] Specifically, for the features , , , Apply cross attention mechanism to get features , , :

[0062]

[0063]

[0064]

[0065] in, for The number of queries, for The number of queries, for The number of queries, , , and , , They are The keys and values ​​of for , , The dimension of is the cross attention mechanism function.

[0066] S22: Input the load voltage correlation feature, the temperature voltage correlation feature and the humidity voltage correlation feature into the MLP network for processing, and obtain the output feature as the actual transformer ratio prediction value.

[0067] In this example, each feature , , Input into the MLP network to get the output features The output features The value is used as the predicted value of the actual transformer ratio .

[0068] S3: Multi-task learning of regression task and classification task is performed on the predicted value of the actual transformation ratio of the transformer to obtain the actual transformation ratio of the transformer.

[0069] The purpose of the regression task is to output the accurate actual transformation ratio value of the transformer, and the purpose of the classification task is to divide the actual transformation ratio value of the transformer into ranges and calculate the measured transformation ratio value. Variance of the data and average , determine the category of the classification task based on the variance and mean, , They represent classification task 1 and classification task 2 respectively. Classification tasks can assist regression tasks. When the classification task is confirmed as classification task 1 or classification task 2, the output value of the regression task will not deviate from the numerical range. , , which promotes the output value of the regression task to be more accurate.

[0070] In some optional embodiments, step S3 includes:

[0071] S31: Use the actual transformer ratio prediction value The activation function obtains the predicted probability value, and the predicted probability value uses the cross entropy loss function for classification task learning.

[0072] Specifically, the output features use The activation function obtains the predicted probability value , indicating the predicted probability value of classification task 1 or classification task 2, and then Use the cross entropy loss function for classification task learning.

[0073] The classification task uses the cross entropy loss function, the formula is as follows:

[0074]

[0075] in, Indicates category; Indicates the number of input samples of the model; is a sign function with a value of 0 or 1. The true category is If yes, it takes 1, otherwise it takes 0; Representation sample Belongs to category The predicted probability of .

[0076] S32: The predicted value of the transformer actual transformation ratio is subjected to regression task learning using the mean square error loss function.

[0077] Specifically, the regression task uses the mean square error loss function, the formula is as follows:

[0078]

[0079] in, represents the model's prediction of the actual transformer ratio, Indicates the measured transformation ratio of the transformer.

[0080] S33: The total model loss is obtained by weighted summing the cross entropy loss of the classification task and the mean square error loss of the regression task.

[0081] Specifically, the total loss function of the model is a weighted sum of the classification task and the regression task, as shown below

[0082]

[0083] in, , is the weight.

[0084] In some optional embodiments, a dynamic weight averaging algorithm is used to adjust the weight coefficients of the regression task and the classification task.

[0085] Since the convergence of different tasks may be different, fixing the weight of a single task loss function will make the model as a whole biased towards a certain task during the learning process, resulting in the model being unable to obtain the generalization features of multiple tasks and unable to reflect the effectiveness of multi-task learning.

[0086] Therefore, the Dynamic Weight Averaging (DWA) algorithm is used to adjust the weight coefficient of the single task loss. This method balances the learning speed of different tasks by focusing on the loss value of a single task. The idea of ​​the DWA algorithm is as follows: and Adjust the weight coefficients of regression and classification tasks , j is 1 or 2, Indicates the number of single tasks. In the current model ; is the number of iterations of the model; Indicates the task exist The loss value at the iteration; Indicates the task exist The training speed at iterations; To set the constant, The larger the value, the more evenly the task weight coefficients are distributed. In this example, Set the value of to 2.

[0087] S34: Iteratively adjust the model parameters in a direction that reduces the total loss of the model to obtain the actual transformation ratio of the transformer.

[0088] Specifically, adjust the parameters of the MLP network and linear network layers as well as the joint self-attention mechanism and cross-attention mechanism.

[0089] S4: The voltage on the high-voltage side of the transformer is calculated based on the actual transformation ratio of the transformer, and the current actual error of the transformer is calculated in combination with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal.

[0090] In this example, the actual transformer ratio obtained by the model is Calculate the voltage on the high voltage side of the transformer , so as to obtain the transformer high voltage side voltage . Collect the low voltage side voltage data of the voltage transformer , through the transformer low voltage side voltage and rated transformation ratio Calculate the rated voltage of the high voltage side of the transformer Then the rated voltage on the high voltage side of the transformer and transformer high voltage side voltage Calculating the ratio error of a transformer .

[0091] Specifically, according to the formula Calculate the voltage on the high voltage side of the transformer ,in, is the voltage on the high voltage side of the transformer, is the actual transformation ratio of the transformer.

[0092] According to the formula Calculate the rated voltage of the high voltage side of the transformer ,in, is the voltage data of the low voltage side, is the rated transformation ratio.

[0093] According to the formula , calculate the actual current error of the transformer.

[0094] Finally, it is determined whether the voltage transformer exceeds its basic error limit according to the ratio error of the voltage transformer and the accuracy grade of the voltage transformer, thereby evaluating the operating state of the voltage transformer.

[0095] On the other hand, the present invention also provides a distribution network voltage transformer state evaluation device, comprising: a data processing module, a prediction value output module, a learning module and a judgment module.

[0096] The data processing module is used to split the endogenous data and then embed the variables, and use the joint self-attention mechanism and fuse it with itself to input the linear network layer to obtain the endogenous intermediate features, and use the linear network layer to embed the variables of the exogenous data to obtain the exogenous intermediate features, where the endogenous data includes the voltage data on the low-voltage side of the transformer, and the exogenous data includes the load data, temperature data and humidity data on the low-voltage side; the predicted value output module is used to apply the cross-attention mechanism to the endogenous intermediate features and the exogenous intermediate features to process them, obtain the internal and external source correlation features, and input them into the MLP network for processing, obtain the output features and use them as the actual transformation ratio prediction value of the transformer; the learning module is used to perform multi-task learning of regression tasks and classification tasks on the actual transformation ratio prediction value of the transformer at the same time, and obtain the actual transformation ratio of the transformer; the judgment module is used to calculate the high-voltage side voltage of the transformer according to the actual transformation ratio of the transformer, and calculate the current actual error of the transformer in combination with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal.

[0097] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the above-described device and each module and unit can refer to the corresponding process in the aforementioned distribution network voltage transformer state assessment method embodiment, and will not be repeated here.

[0098] The experimental results of the voltage transformer based on the present invention are shown in Tables 1 and 2 below. The error evaluation value obtained by the traditional offline inspection method is compared with the error evaluation value obtained by the method provided by the present invention, and the mean absolute error MAE, mean absolute percentage error MAPE, and root mean square error RMSE are selected as evaluation indicators. The experimental data are collected from the voltage transformer in the actual distribution network, the sampling number of the voltage transformer is 8, and the comparison item is the ratio difference of the voltage transformer.

[0099] Table 1

[0100]

[0101] Table 2

[0102]

[0103] In summary, this scheme divides the endogenous data and then embeds the variables, uses the joint self-attention mechanism and fuses it with itself to input the linear network layer to obtain the endogenous intermediate features, and uses the linear network layer to embed the variables of the exogenous data to obtain the exogenous intermediate features respectively; the cross-attention mechanism is applied to the endogenous intermediate features and the exogenous intermediate features to obtain the endogenous and exogenous correlation features, and input them into the MLP network for processing, and the output features are obtained as the actual transformer ratio prediction value; the transformer actual ratio prediction value is simultaneously subjected to multi-task learning of regression tasks and classification tasks to obtain the actual transformer ratio; the transformer high-voltage side voltage is calculated according to the actual transformer ratio, and the current actual transformer error is calculated in combination with the rated voltage of the transformer high-voltage side to determine whether the current error state of the voltage transformer is abnormal. The transformer data is divided into endogenous data and exogenous data, the internal dependency of the endogenous data is extracted, and the coupling relationship between the exogenous data and the endogenous data is calculated as a whole, and the endogenous data and the exogenous data are combined to improve the accuracy of the model. By segmenting the endogenous data and applying a joint self-attention mechanism to learn the internal associations of the endogenous data, the amount of computation is reduced and the computational efficiency of the model is improved.

[0104] In the description of the present application, it should be noted that the terms "upper", "lower", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying 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 application. 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 an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0105] It should be noted that, in this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0106] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A method for evaluating the state of a voltage transformer in a distribution network, characterized in that: The following steps are involved: The endogenous data is segmented and then embedded into variables. The endogenous intermediate features are obtained by using the joint self-attention mechanism and fusing it with itself into the linear network layer. The exogenous data is embedded into variables using the linear network layer to obtain the exogenous intermediate features, including: The transformer low-voltage side voltage data is evenly divided to obtain low-voltage side voltage division data; The low-voltage side voltage segmentation data is subjected to the joint self-attention mechanism and fused with its own input linear network layer to obtain the intermediate features of the low-voltage side voltage, including: Slice data for low voltage side voltage Use linear network layers to embed variables , for Using linear layers to embed variables The corresponding data, among which, , is the kth segment after the transformer low-voltage side voltage data is evenly segmented; right Apply joint self-attention mechanism to obtain transition features : in, is the activation function, represents transpose, for The number of queries, , They are The keys and values ​​of for Dimensions; for The number of queries, , They are The keys and values ​​of for Dimensions; , is the self-attention mechanism function, atten_k-1(...,atten_2(atten_1(,)),)...,) joint self-attention mechanism function; The characteristics With features After fusion, the linear layer is input to obtain the intermediate characteristics of the low-voltage side voltage ; The low-voltage side load data, temperature data and humidity data are embedded using a linear network layer to obtain the low-voltage side load intermediate features, temperature intermediate features and humidity intermediate features respectively; among which, the endogenous data includes the transformer low-voltage side voltage data, and the exogenous data includes the low-voltage side load data, temperature data and humidity data; Applying a cross-attention mechanism to the endogenous intermediate features and the exogenous intermediate features to obtain endogenous and exogenous correlation features, and inputting them into the MLP network for processing to obtain output features as the actual transformer ratio prediction value, including: applying a cross-attention mechanism to the low-voltage side voltage intermediate features and the low-voltage side load intermediate features, temperature intermediate features and humidity intermediate features to obtain load voltage correlation features, temperature voltage correlation features and humidity voltage correlation features; inputting the load voltage correlation features, temperature voltage correlation features and humidity voltage correlation features into the MLP network for processing to obtain output features as the actual transformer ratio prediction value; Multi-task learning of regression task and classification task is performed on the predicted value of the actual transformer ratio to obtain the actual transformer ratio; The voltage on the high-voltage side of the transformer is calculated based on the actual transformation ratio of the transformer, and the current actual error of the transformer is calculated in combination with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal.

2. The method for evaluating the state of a voltage transformer in a distribution network according to claim 1, characterized in that: The multi-task learning of simultaneously performing regression task and classification task on the predicted value of the actual transformation ratio of the transformer to obtain the actual transformation ratio of the transformer includes: Use the actual transformer ratio prediction value The activation function obtains the predicted probability value, and the predicted probability value uses the cross entropy loss function for classification task learning; The predicted value of the transformer actual transformation ratio is used for regression task learning using the mean square error loss function; The total model loss is obtained by weighted summing the cross entropy loss of the classification task and the mean square error loss of the regression task; The model parameters are adjusted iteratively in the direction of reducing the total loss of the model to obtain the actual transformation ratio of the transformer.

3. The method for evaluating the state of a voltage transformer in a distribution network according to claim 2, characterized in that: A dynamic weight averaging algorithm is used to adjust the weight coefficients of regression and classification tasks.

4. The method for evaluating the state of a voltage transformer in a distribution network according to claim 3, characterized in that: according to and Adjust the weight coefficients of regression and classification tasks , Indicates the number of single tasks. In the current model ; is the number of iterations of the model; Indicates the task exist The loss value at the iteration; Indicates the task exist The training speed at iterations; To set a constant.

5. The method for evaluating the state of a voltage transformer in a distribution network according to claim 1, characterized in that: According to the formula Calculate the voltage on the high voltage side of the transformer ,in, is the voltage on the high voltage side of the transformer, is the actual transformation ratio of the transformer.

6. The method for evaluating the state of a voltage transformer in a distribution network according to claim 5, characterized in that: According to the formula Calculate the rated voltage of the high voltage side of the transformer ,in, is the voltage data of the low voltage side, is the rated transformation ratio.

7. The method for evaluating the state of a voltage transformer in a distribution network according to claim 6, characterized in that: According to the formula , calculate the actual error of the current transformer E .

8. A distribution network voltage transformer status assessment device, characterized in that: include: The data processing module is used to segment the endogenous data and then embed the variables, and use the joint self-attention mechanism to fuse the input linear network layer with itself to obtain the endogenous intermediate features, and use the linear network layer to embed the variables of the exogenous data to obtain the exogenous intermediate features, including: The transformer low-voltage side voltage data is evenly divided to obtain low-voltage side voltage division data; The low-voltage side voltage segmentation data is subjected to the joint self-attention mechanism and fused with its own input linear network layer to obtain the intermediate features of the low-voltage side voltage, including: Slice data for low voltage side voltage Use linear network layers to embed variables , for Using linear layers to embed variables The corresponding data, among which, , is the kth segment after the transformer low-voltage side voltage data is evenly segmented; right Apply joint self-attention mechanism to obtain transition features : in, is the activation function, represents transpose, for The number of queries, , They are The keys and values ​​of for Dimensions; for The number of queries, , They are The keys and values ​​of for Dimensions; , is the self-attention mechanism function, atten_k-1(...,atten_2(atten_1(,)),)...,) joint self-attention mechanism function; The characteristics With features After fusion, the linear layer is input to obtain the intermediate characteristics of the low-voltage side voltage ; The low-voltage side load data, temperature data and humidity data are embedded using a linear network layer to obtain the low-voltage side load intermediate features, temperature intermediate features and humidity intermediate features respectively; among which, the endogenous data includes the transformer low-voltage side voltage data, and the exogenous data includes the low-voltage side load data, temperature data and humidity data; A prediction value output module is used to apply a cross-attention mechanism to the endogenous intermediate features and the exogenous intermediate features for processing, obtain the endogenous and exogenous correlation features, and input them into the MLP network for processing to obtain the output features as the actual transformer ratio prediction value, including: applying a cross-attention mechanism to the low-voltage side voltage intermediate features and the low-voltage side load intermediate features, the temperature intermediate features and the humidity intermediate features for processing to obtain the load voltage correlation features, the temperature voltage correlation features and the humidity voltage correlation features; inputting the load voltage correlation features, the temperature voltage correlation features and the humidity voltage correlation features into the MLP network for processing to obtain the output features as the actual transformer ratio prediction value; A learning module, which is used to simultaneously perform multi-task learning of regression tasks and classification tasks on the predicted value of the actual transformation ratio of the transformer to obtain the actual transformation ratio of the transformer; The judgment module is used to calculate the voltage on the high-voltage side of the transformer according to the actual transformation ratio of the transformer, and calculate the current actual error of the transformer in combination with the rated voltage on the high-voltage side of the transformer to determine whether the current error state of the voltage transformer is abnormal.

Images