Discharge voltage prediction method and system for air gap in converter valve module

Abstract

The invention discloses a method and a system for predicting discharge voltage of air gaps in a converter valve module. The method comprises the following steps: acquiring discharge voltage data of air gaps in multiple groups of converter valve modules and performing normalization processing; dividing a hyper-parameter training set for hyper-parameter adjustment and optimization from the multiple groups of converter valve module internal air gap discharge voltage data, and performing hyper-parameter adjustment and optimization of a LightGBM model based on the hyper-parameter training set to determine an optimal hyper-parameter; establishing a LightGBM model based on the optimal hyper-parameter, training based on a model training set divided from the internal air gap discharge voltage data of the plurality of groups of converter valve modules, and determining internal parameters of the LightGBM model so as to determine a discharge voltage prediction model; and key structure parameters and meteorological characteristics of the converter valve module of the demand structure are selected and input to the discharge voltage prediction model, and a discharge voltage prediction value of the gap of the converter valve module of the demand structure is acquired.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of high voltage and insulation technology, and more particularly, to a discharge voltage prediction method and system for an air gap inside a converter valve module. BACKGROUND

[0002] The current energy pattern of "west-to-east and north-to-south" makes the DC power transmission project a key support for the national energy strategy. The UHV DC power transmission has the technical advantages of large transmission capacity (a single line can reach 12 million kilowatts), long distance (the longest is more than 3000 kilometers), and low loss (only 3% per kilometer), which can efficiently solve the new energy consumption problem in the "three north" region and promote the development of the western clean energy base.

[0003] The converter valve is the core equipment of the DC power transmission project, mainly responsible for the key function of AC-DC power conversion, and is the technical basis of UHV DC power transmission, directly affecting the transmission efficiency and reliability of the project, and playing a decisive role in realizing new energy cross-regional consumption and power grid interconnection. Currently, the design of the air gap inside the converter valve module mainly refers to GB / T 16935.1-2008 "Insulation Coordination of Electrical Equipment in Low-Voltage Systems Part 1: Principles and Tests", but the insulation design method provided by this standard is not for the converter valve module, which often leads to excessive insulation design margin of the converter valve, limited applicability, and limited altitude range. At present, a large number of discharge characteristic tests have been carried out on the internal gap of the converter valve module in different altitudes, but there is still a lack of a suitable method to guide the engineering design.

[0004] Therefore, a discharge voltage prediction method for the air gap inside the converter valve module is needed. SUMMARY

[0005] The present application provides a discharge voltage prediction method and system for the air gap inside the converter valve module to solve the problem of how to efficiently determine the discharge voltage of the air gap inside the converter valve module.

[0006] To solve the above problems, according to one aspect of the present application, a discharge voltage prediction method for the air gap inside the converter valve module is provided, which comprises:

[0007] Obtain multiple sets of discharge voltage data of the air gap inside the converter valve module and perform normalization processing; wherein each set of discharge voltage data of the air gap inside the converter valve module includes discharge voltage, key structural parameters and meteorological characteristics;

[0008] Divide a hyperparameter training set for hyperparameter tuning from the multiple sets of discharge voltage data of the air gap inside the converter valve module, and perform LightGBM model hyperparameter tuning based on the hyperparameter training set to determine the optimal hyperparameters;

[0009] establishing a LightGBM model based on the optimal hyperparameters, and training the LightGBM model based on a model training set divided from the multiple sets of converter valve module internal air gap discharge voltage data, to determine internal parameters of the LightGBM model, thereby determining the discharge voltage prediction model;

[0010] inputting key structural parameters of the converter valve module of the demand structure and meteorological features into the discharge voltage prediction model, to obtain a predicted value of the discharge voltage of the gap of the converter valve module of the demand structure.

[0011] Preferably, the key structural parameters include gap structure and gap distance, and the meteorological features include dry temperature, wet temperature, relative humidity, and air pressure.

[0012] Preferably, the method normalizes each set of converter valve module internal air gap discharge voltage data by using a Min-Max normalization method.

[0013] Preferably, a hyperparameter training set for hyperparameter tuning is divided from the multiple sets of converter valve module internal air gap discharge voltage data, and hyperparameter tuning of the LightGBM model is performed based on the hyperparameter training set to determine optimal hyperparameters, including:

[0014] The multiple sets of converter valve module internal air gap discharge voltage data are divided into a model training set and a hyperparameter training set by using a k-fold cross-validation method.

[0015] The hyperparameters of the LightGBM are configured, the gradient decision boosting tree (LGBT) is selected as an enhanced method, and the learning rate (lr), the number of enhanced integrations (n_boost), the number of leaves of decision tree growth (n_leafs), the maximum depth (depth), and the minimum segmentation gain (min_gain) are selected as hyperparameters that need to be tuned.

[0016] According to the requirements of the model prediction task, the mean square error (MSE) of the output result of the LightGBM model is used as an optimization objective function:

[0017]

[0018] where y i is an actual value, is a predicted value, and n is the number of samples. By minimizing the objective function, the hyperparameter training set is tested, appropriate hyperparameter grid points are selected, and the MSE of the optimal hyperparameters is found by iteration, so as to obtain the optimal hyperparameters of the LightGBM.

[0019] Preferably, the internal parameters of the LightGBM model are determined based on a model training set divided from the plurality of sets of internal air gap discharge voltage data of the converter valve module, so as to determine the discharge voltage prediction model, comprising:

[0020] The plurality of sets of internal air gap discharge voltage data of the converter valve module are selected as the model training set, and the key structural parameters and meteorological features are selected as the feature input;

[0021] The mean square error (MSE) is used as the optimization objective function, the first-order and second-order regularization parameters λ1 and λ2 are both set to 0.1, and the training is performed to iteratively generate n_boost decision trees, the internal parameters of each decision tree are trained, each subsequent tree is fitted to the residual of the previous tree, and the discharge voltage prediction model is obtained based on the weighted combination of the plurality of trees;

[0022] Wherein, the splitting gain Gain of the decision tree during training is determined by the following formula:

[0023]

[0024] Wherein, G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H L is the second-order gradient of the left tree, and H R is the second-order gradient of the right tree.

[0025] The training process is represented as:

[0026]

[0027] Wherein, is the prediction value of the t-th round, α i is the weight of the i-th tree, I(x∈R i ) is an indicator function, which takes a value of 1 when x belongs to the region R i of the i-th tree, and 0 otherwise.

[0028] Preferably, the method further comprises:

[0029] The discharge voltage prediction value of the converter valve module gap of the demand structure is compared with the discharge voltage design demand value to determine the insulation margin prediction value.

[0030] According to another aspect of the present application, a discharge voltage prediction system for an internal air gap of a converter valve module is provided, the system comprising:

[0031] The data processing unit is configured to obtain and normalize a plurality of sets of internal air gap discharge voltage data of the converter valve modules; wherein each set of internal air gap discharge voltage data of the converter valve modules comprises a discharge voltage, key structural parameters, and meteorological characteristics.

[0032] The optimal hyperparameter determination unit is configured to divide a hyperparameter training set for hyperparameter tuning from the plurality of sets of internal air gap discharge voltage data of the converter valve modules, and perform LightGBM model hyperparameter tuning based on the hyperparameter training set to determine optimal hyperparameters.

[0033] The model determination unit is configured to establish a LightGBM model based on the optimal hyperparameters, and train the model based on a model training set divided from the plurality of sets of internal air gap discharge voltage data of the converter valve modules to determine internal parameters of the LightGBM model, thereby determining a discharge voltage prediction model.

[0034] The voltage prediction unit is configured to input key structural parameters and meteorological characteristics of a converter valve module of a required structure into the discharge voltage prediction model to obtain a discharge voltage prediction value of a gap of the converter valve module of the required structure.

[0035] Preferably, the key structural parameters comprise a gap structure and a gap distance, and the meteorological characteristics comprise a dry temperature, a wet temperature, a relative humidity, and an air pressure.

[0036] Preferably, the data processing unit normalizes each set of internal air gap discharge voltage data of the converter valve modules by using a Min-Max normalization system.

[0037] Preferably, the optimal hyperparameter determination unit divides a hyperparameter training set for hyperparameter tuning from the plurality of sets of internal air gap discharge voltage data of the converter valve modules, and performs LightGBM model hyperparameter tuning based on the hyperparameter training set to determine optimal hyperparameters, comprising:

[0038] The plurality of sets of internal air gap discharge voltage data of the converter valve modules are divided into a model training set and a hyperparameter training set by using a k-fold cross-validation system.

[0039] The hyperparameters of LightGBM are configured, the gradient decision boosting tree (LGBT) is selected as an enhanced system, the learning rate (lr), the number of enhanced integrations (n_boost), the number of leaves of decision tree growth (n_leafs), the maximum depth (depth), and the minimum segmentation gain (min_gain) are selected as hyperparameters to be tuned.

[0040] According to the requirements of a model prediction task, the mean square error (MSE) of a LightGBM model output result is used as an optimization objective function:

[0041]

[0042] where y i is the actual value, is the predicted value, n is the number of samples, by minimizing the objective function, the hyperparameter training set is tested, the appropriate hyperparameter grid point is selected, the solution is traversed, and the hyperparameter with the minimum MSE is found as the optimal hyperparameter of LightGBM.

[0043] Preferably, wherein the model determination unit is trained based on the model training set divided from the plurality of groups of converter valve module internal air gap discharge voltage data, determines the internal parameters of the LightGBM model, and determines the discharge voltage prediction model, comprising:

[0044] The discharge voltage data of the internal air gap of the plurality of groups of converter valve modules is used as the model training set, and the key structural parameters and meteorological features are selected as the feature input;

[0045] The mean square error MSE is used as the optimization objective function, the first-order and second-order regularization parameters λ1 and λ2 are set to 0.1, and training is performed to generate n_boost decision trees, the internal parameters of each decision tree are trained, each subsequent tree fits the residual of the previous tree, and the discharge voltage prediction model is obtained based on the weighted combination of multiple trees;

[0046] wherein the split gain Gain of the decision tree during training is determined by the following formula:

[0047]

[0048] where G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H L is the second-order gradient of the left tree, and H R is the second-order gradient of the right tree.

[0049] The training process is represented as:

[0050]

[0051] where y i is the predicted value of the t-th round, α i is the weight of the i-th tree, and I(x∈R i ) is an indicator function that takes the value 1 when x belongs to the region R i of the i-th tree, and 0 otherwise.

[0052] Preferably, wherein the system further comprises:

[0053] The insulation margin determination unit is configured to compare the discharge voltage prediction value of the gap of the converter valve module of the demand structure with the discharge voltage design demand value, so as to determine an insulation margin prediction value.

[0054] The present application provides a discharge voltage prediction method and system for the internal air gap of a converter valve module, comprising: obtaining and normalizing a plurality of sets of discharge voltage data for the internal air gap of a converter valve module; wherein each set of discharge voltage data for the internal air gap of a converter valve module comprises a discharge voltage, key structure parameters, and meteorological characteristics; dividing a hyperparameter training set for hyperparameter tuning from the plurality of sets of discharge voltage data for the internal air gap of a converter valve module, and performing hyperparameter tuning of a LightGBM model based on the hyperparameter training set to determine optimal hyperparameters; establishing a LightGBM model based on the optimal hyperparameters, and training the model based on a model training set divided from the plurality of sets of discharge voltage data for the internal air gap of a converter valve module to determine internal parameters of the LightGBM model, thereby determining a discharge voltage prediction model; inputting key structure parameters and meteorological characteristics of a converter valve module of a demand structure into the discharge voltage prediction model to obtain a discharge voltage prediction value for the gap of the converter valve module of the demand structure. The present application can predict the discharge voltage of the internal air gap of a converter valve module under different meteorological conditions without the need for complicated calculation procedures, is more convenient, and has a wider altitude coverage range, can predict the discharge voltage under extreme meteorological conditions at high altitudes, and provides a reference for subsequent optimization design of the internal air gap distance of a converter valve module. BRIEF DESCRIPTION OF DRAWINGS

[0055] The exemplary embodiments of the present application can be more completely understood in reference to the following drawings:

[0056] Figure 1 A flowchart of the discharge voltage prediction method 100 for the internal air gap of a converter valve module according to an embodiment of the present application;

[0057] Figure 2 A LightGBM model training flowchart according to an embodiment of the present application;

[0058] Figure 3 A structure diagram of the discharge voltage prediction system 300 for the internal air gap of a converter valve module according to an embodiment of the present application. DETAILED DESCRIPTION

[0059] Reference will now be made to the drawings to describe the exemplary embodiments of the present application in greater detail. The present application can be variously embodied and is not limited to the embodiments described herein, which are provided for the purposes of disclosure and to fully and completely disclose the present application to those skilled in the art. The terms used in the exemplary embodiments of the present application shown in the drawings are terms that are used to describe the present application and are not intended to limit the present application. In the drawings, the same elements are denoted by the same reference numerals.

[0060] Unless otherwise defined, the terms (including technical terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. In addition, it is to be understood that the terms defined by commonly used dictionaries are to be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and are not to be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0061] Figure 1 A flowchart of the discharge voltage prediction method 100 for the internal air gap of the converter valve module according to the embodiments of the present application. As shown in Figure 1 The discharge voltage prediction method for the internal air gap of the converter valve module provided by the embodiments of the present application can predict the discharge voltage of the internal air gap of the converter valve module under different weather conditions without complicated calculation procedures, is more convenient, and has a wider altitude coverage range, and can predict the discharge voltage under extreme weather conditions at high altitudes, thereby providing a reference for subsequent optimization design of the internal air gap distance of the converter valve module. The discharge voltage prediction method 100 for the internal air gap of the converter valve module provided by the embodiments of the present application starts from step 101, in which a plurality of sets of internal air gap discharge voltage data of the converter valve module are obtained and normalized in step 101. Each set of internal air gap discharge voltage data of the converter valve module includes a discharge voltage, key structural parameters, and weather characteristics.

[0062] Preferably, the key structural parameters include gap structure and gap distance, and the weather characteristics include dry temperature, wet temperature, relative humidity, and air pressure.

[0063] Preferably, the method uses the Min-Max normalization method to normalize each set of internal air gap discharge voltage data of the converter valve module.

[0064] In the present application, a total of 202 sets of internal air gap discharge voltage data of the converter valve module are obtained based on experiments, the gap structure and the gap distance are selected as the key structural parameters, the dry temperature, the wet temperature, the relative humidity, and the air pressure are selected as the weather characteristics, the structural parameters and the weather characteristics are used as the feature input of the training model, and the Min-Max normalization method is used to process the data.

[0065] Specifically, according to the 202 pieces of discharge voltage data of the air gap inside the converter valve module at different altitudes, the gap structure and the gap distance are selected as the key structural parameters, and the dry temperature, the wet temperature, the relative humidity and the air pressure are selected as the meteorological characteristics, and the structural parameters and the meteorological characteristics are taken as the feature inputs of the training model.

[0066] Among them, the Min-Max normalization method is adopted to process the data, and the formula is:

[0067]

[0068] Among them, x i is the original feature value, x min and x max are the minimum and maximum values of the feature. Through normalization, all feature values are scaled to the range of 0 to 1.

[0069] In step 102, the hyperparameter training set for hyperparameter tuning is divided from the plurality of sets of discharge voltage data of the air gap inside the converter valve module, and the LightGBM model hyperparameter tuning is performed based on the hyperparameter training set to determine the optimal hyperparameters.

[0070] Preferably, wherein the hyperparameter training set for hyperparameter tuning is divided from the plurality of sets of discharge voltage data of the air gap inside the converter valve module, and the LightGBM model hyperparameter tuning is performed based on the hyperparameter training set to determine the optimal hyperparameters, comprising:

[0071] The plurality of sets of discharge voltage data of the air gap inside the converter valve module are divided into a model training set and a hyperparameter training set by using the k-fold cross-validation method;

[0072] The hyperparameters of LightGBM are configured, the gradient decision boosting tree LGBT is selected as the enhancement method, and the learning rate lr, the number of enhanced integrations n_boost, the number of leaves of decision tree growth n_leafs, the maximum depth depth and the minimum segmentation gain min_gain are selected as the hyperparameters to be tuned;

[0073] According to the requirements of the model prediction task, the mean square error (MSE) of the LightGBM model output result is taken as the optimization objective function:

[0074]

[0075] Among them, y i is the actual value, is the predicted value, and n is the sample number. By minimizing the objective function, the hyperparameter training set is tested, the appropriate hyperparameter grid points are selected, the solution is iterated, and the hyperparameters with the minimum MSE are found as the optimal hyperparameters of LightGBM.

[0076] In the present application, the data set for hyperparameter tuning is divided from the discharge voltage data of the air gap inside the converter valve module. The mean square error (MSE) of the LightGBM model output result is taken as the optimization objective function, the gradient decision boosting tree (LGBT) is selected as the enhancement method, and the learning rate lr, the number of enhanced integrations n_boost, the number of leaves of decision tree growth n_leafs, the maximum depth depth, and the minimum segmentation gain min_gain are selected as the hyperparameters to be tuned. The hyperparameter optimization set is optimized, the appropriate hyperparameter grid points are selected for iteration, and the hyperparameters with the minimum MSE are selected as the optimal hyperparameters of LightGBM.

[0077] In combination Figure 2 As shown, specifically, the process of determining the optimal hyperparameters includes:

[0078] 1. Select the data for hyperparameter tuning

[0079] From the 202 discharge voltage data of the air gap inside the converter valve module, the data set for hyperparameter tuning is divided. The k-fold cross-validation method is used to divide the data set into 5 folds, and the data set is divided into a model training set and a hyperparameter training set for model parameter optimization and verification.

[0080] 2. Select the hyperparameters of LightGBM to be tuned

[0081] Configure the hyperparameters of LightGBM, select the gradient decision boosting tree (LGBT) as the enhancement method, and select the learning rate lr, the number of enhanced integrations n_boost, the number of leaves of decision tree growth n_leafs, the maximum depth depth, and the minimum segmentation gain min_gain as the hyperparameters to be tuned.

[0082] 3. Set the optimization objective function

[0083] According to the requirements of the model prediction task, the optimization objective function is set. The mean square error (MSE) of the LightGBM model output result is taken as the optimization objective function:

[0084]

[0085] Where y i is the actual value, is the predicted value, and n is the sample size. By minimizing the objective function, the hyperparameter optimization set is tested, the appropriate hyperparameter grid points are selected for iteration, and the hyperparameters with the minimum MSE are selected as the optimal hyperparameters of LightGBM.

[0086] At step 103, a LightGBM model is established based on the optimal hyperparameters, and the internal parameters of the LightGBM model are determined based on the model training set divided from the plurality of groups of thyristor module internal air gap discharge voltage data, so as to determine the discharge voltage prediction model.

[0087] Preferably, wherein the internal parameters of the LightGBM model are determined based on the model training set divided from the plurality of groups of thyristor module internal air gap discharge voltage data, so as to determine the discharge voltage prediction model, comprising:

[0088] From the plurality of groups of thyristor module internal air gap discharge voltage data as the model training set, the key structural parameters and the meteorological features are selected as the feature input;

[0089] The mean square error MSE is taken as the optimization objective function, the first-order and second-order regularization parameters λ1 and λ2 are set to 0.1, and training is performed to iteratively generate n_boost decision trees, the internal parameters of each decision tree are trained, each subsequent tree is fitted to the residual of the previous tree, and the discharge voltage prediction model is obtained based on the weighted combination of multiple trees;

[0090] Wherein, the split gain Gain of the decision tree during training is determined by the following formula:

[0091]

[0092] Wherein, G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H L is the second-order gradient of the left tree, and H R is the second-order gradient of the right tree.

[0093] The training process is represented as:

[0094]

[0095] Wherein, is the prediction value of the tth round, α i is the weight of the ith tree, I(x∈R i ) is an indicator function, which takes a value of 1 when x belongs to the region R i of the ith tree, and 0 otherwise.

[0096] In the present application, according to the optimal hyperparameters obtained in step two, a LightGBM model is established, the air gap discharge voltage data of the converter valve module is selected as the model training set data from the data set divided in step two, the gap structure and the gap distance are selected as the key structure parameters, the dry temperature, the wet temperature, the relative humidity and the air pressure are selected as the meteorological characteristics as the characteristic input, and the LightGBM discharge voltage prediction model is trained.

[0097] In the present application, according to the optimal hyperparameters obtained in step 102, a LightGBM model is established, the air gap discharge voltage data of the converter valve module is selected as the model training set data from the data set divided in step 102, the gap structure and the gap distance are selected as the key structure parameters, the dry temperature, the wet temperature, the relative humidity and the air pressure are selected as the meteorological characteristics as the characteristic input, and the LightGBM discharge voltage prediction model is trained.

[0098] In combination Figure 2 As shown, specifically, the process of training the LightGBM model includes:

[0099] 1. Setting the LightGBM parameters and establishing the LightGBM model:

[0100] According to the optimal hyperparameters obtained in step 102, the LightGBM model is established, and the specific parameters include the learning rate (learning rate), the number of enhanced integrations (n_boost), the number of leaves of the decision tree growth (n_leaves), the maximum depth (depth) and the minimum segmentation gain (min_gain).

[0101] 2. Selecting the model training set data:

[0102] The air gap discharge voltage data of the converter valve module is selected as the model training set data from the normalized data set, and the gap structure and the gap distance of step one are selected as the structure parameters and the air pressure, the dry temperature and the relative humidity as the meteorological condition parameters as the characteristic input.

[0103] 3. Training the LightGBM model:

[0104] The mean square error (MSE) is used as the optimization objective function, the first-order and second-order regularization parameters λ1 and λ2 are set to 0.1, and the splitting gain Gain of the decision tree during training is determined by the following formula:

[0105]

[0106] Wherein, G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H LH is the second order gradient of the left tree. R H is the second order gradient of the right tree.

[0107] n_boost decision trees are iteratively generated, internal parameters of each decision tree are trained, each subsequent tree is fitted to the residual of the previous tree, and a discharge voltage prediction model is obtained based on the weighted combination of multiple trees. The training process can be represented as:

[0108]

[0109] wherein, is the prediction value of the tth round, and i is the weight of the ith tree, and i is an indicator function, which takes a value of 1 when x belongs to the region R i of the ith tree, and 0 otherwise.

[0110] In step 104, the key structural parameters of the converter valve module of the demand structure and the meteorological characteristics are selected as inputs to the discharge voltage prediction model, and the discharge voltage prediction value of the gap of the converter valve module of the demand structure is obtained.

[0111] In the present application, according to the LightGBM model obtained in step 103, the appropriate gap structure can be selected according to the engineering requirements, the gap distance and the meteorological parameters are input into the LightGBM discharge voltage prediction model, and the prediction value of the discharge voltage of the valve module gap of the corresponding structure is obtained. The LightGBM discharge voltage prediction model can evaluate the insulation margin, according to the actual design clearance of the air gap inside the converter valve module, the actual discharge voltage is predicted, compared with the existing engineering design requirements, and the prediction value of the insulation margin is given.

[0112] As shown in Figure 2 , specifically, the process of discharge voltage prediction includes:

[0113] 1. Determine the structure parameters and meteorological condition parameters of the data to be predicted:

[0114] According to the experimental design or actual application requirements, the specific structural parameters of the internal air gap of the converter valve module, such as gap distance and gap structure, are determined, which need to be predicted for discharge voltage. Determine the related meteorological condition parameters, such as dry temperature, wet temperature, relative humidity and air pressure, as the feature input of the model.

[0115] 2. Input the data into the trained LightGBM model to obtain the prediction value of the discharge voltage, and the specific formula is as follows:

[0116]

[0117] Wherein, x is a characteristic determined by a structure parameter of the valve structure and a meteorological condition parameter.

[0118] Preferably, wherein the method further comprises:

[0119] The discharge voltage prediction value of the converter valve module gap of the demand structure is compared with the discharge voltage design demand value to determine the insulation margin prediction value.

[0120] The LightGBM discharge voltage prediction model of the application is user-friendly, and has a large amount of test data as the support, only needs the structure parameters of the converter valve module, such as the gap structure, the gap distance, and the air pressure, the dry temperature, and the wet temperature as the related meteorological condition parameters input, and the prediction value of the discharge voltage of the internal air gap of the converter valve module can be obtained. The prediction of the insulation margin can also be given for the existing gap distance design value, including: comparing the predicted result with the actual engineering design to verify, and giving the prediction of the insulation margin.

[0121] The discharge voltage prediction method of the internal air gap of the converter valve module provided by the application trains an intelligent model based on a certain amount of discharge test data of the typical gap inside the converter valve module, trains the discharge voltage intelligent model suitable for the typical complex air gap inside the converter valve module, and can realize the discharge voltage prediction of different types of complex air gaps inside the converter valve module. The method has high prediction performance with fast training speed, can predict the discharge voltage according to different altitudes and different meteorological conditions, fills the blank of the current converter valve module gap insulation design, and has great significance for the design of the converter valve module.

[0122] Figure 3 A structure diagram of the discharge voltage prediction system 300 of the internal air gap of the converter valve module according to the embodiment of the application is shown. Figure 3 As shown in the figure, the discharge voltage prediction system 300 of the internal air gap of the converter valve module provided by the embodiment of the application includes: a data processing unit 301, an optimal hyperparameter determination unit 302, a model determination unit 303, and a voltage prediction unit 304.

[0123] Preferably, the data processing unit 301 is used to acquire and normalize a plurality of groups of discharge voltage data of the internal air gap of the converter valve module; wherein each group of discharge voltage data of the internal air gap of the converter valve module includes: a discharge voltage, a key structure parameter, and a meteorological feature.

[0124] Preferably, wherein the key structure parameter includes: a gap structure and a gap distance; and the meteorological feature includes: a dry temperature, a wet temperature, a relative humidity, and an air pressure.

[0125] Preferably, the data processing unit 301 normalizes each group of converter valve module internal air gap discharge voltage data using a Min-Max normalization system.

[0126] Preferably, the optimal hyperparameter determination unit 302 is configured to divide a hyperparameter training set for hyperparameter tuning from the plurality of groups of converter valve module internal air gap discharge voltage data, and perform LightGBM model hyperparameter tuning based on the hyperparameter training set to determine the optimal hyperparameters.

[0127] Preferably, the optimal hyperparameter determination unit 302 divides a hyperparameter training set for hyperparameter tuning from the plurality of groups of converter valve module internal air gap discharge voltage data, and performs LightGBM model hyperparameter tuning based on the hyperparameter training set to determine the optimal hyperparameters, including:

[0128] The plurality of groups of converter valve module internal air gap discharge voltage data are divided into a model training set and a hyperparameter training set using a k-fold cross-validation system;

[0129] The hyperparameters of LightGBM are configured, the gradient decision boosting tree LGBT is selected as the enhancement system, and the learning rate lr, the number of enhanced integrations n_boost, the number of leaves of decision tree growth n_leafs, the maximum depth depth, and the minimum segmentation gain min_gain are selected as the hyperparameters to be tuned;

[0130] According to the requirements of the model prediction task, the mean square error (MSE) of the LightGBM model output result is taken as the optimization objective function:

[0131]

[0132] where y i is the actual value, is the predicted value, and n is the number of samples. By minimizing the objective function, the hyperparameter training set is tested, the appropriate hyperparameter grid points are selected, and the MSE is solved to find the optimal hyperparameters of LightGBM.

[0133] Preferably, the model determination unit 303 is configured to establish a LightGBM model based on the optimal hyperparameters, train the model based on the model training set divided from the plurality of groups of converter valve module internal air gap discharge voltage data, determine the internal parameters of the LightGBM model, and thus determine the discharge voltage prediction model.

[0134] Preferably, the model determination unit 303 determines the internal parameters of the LightGBM model based on a model training set divided from the plurality of groups of converter valve module internal air gap discharge voltage data, so as to determine the discharge voltage prediction model, comprising:

[0135] The plurality of groups of converter valve module internal air gap discharge voltage data are taken as the model training set, and the key structural parameters and meteorological features are selected as the feature inputs;

[0136] The mean square error MSE is taken as the optimization objective function, the first-order and second-order regularization parameters λ1 and λ2 are both set to 0.1, and training is performed to iteratively generate n_boost decision trees, the internal parameters of each decision tree are trained, each subsequent tree is fitted to the residual of the previous tree, and the discharge voltage prediction model is obtained based on the weighted combination of the plurality of trees;

[0137] Wherein, the splitting gain Gain of the decision tree during training is determined by the following formula:

[0138]

[0139] Wherein, G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H L is the second-order gradient of the left tree, and H R is the second-order gradient of the right tree;

[0140] The training process is represented as:

[0141]

[0142] Wherein, is the prediction value of the t-th round, α i is the weight of the i-th tree, I(x∈R i ) is an indicator function, which takes a value of 1 when x belongs to the region R i of the i-th tree, and 0 otherwise.

[0143] Preferably, the voltage prediction unit 304 is configured to input the key structural parameters and meteorological features of the converter valve module of the demand structure into the discharge voltage prediction model, so as to obtain the discharge voltage prediction value of the gap of the converter valve module of the demand structure.

[0144] Preferably, the system further comprises:

[0145] An insulation margin determination unit configured to compare the discharge voltage prediction value of the gap of the converter valve module of the demand structure with a discharge voltage design requirement value, so as to determine an insulation margin prediction value.

[0146] The discharge voltage prediction system 300 of the internal air gap of the converter valve module of the embodiment of the present application corresponds to the discharge voltage prediction method 100 of the internal air gap of the converter valve module of another embodiment of the present application, which will not be described here again.

[0147] The present application has been described by reference to a few embodiments. However, one of ordinary skill in the art will appreciate that other embodiments, apart from those explicitly disclosed herein, are equally within the scope of the present application.

[0148] Generally, all terms used in the present application are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise herein. All references to "a" or "an" means "at least one" unless otherwise clearly indicated by the context of the disclosure. The steps of any methods disclosed herein need not be performed in the exact order disclosed, unless explicitly stated.

[0149] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0150] The present application is described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing system or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0151] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0152] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are generated to realize the computer-implemented processes in the computer or other programmable devices, and the instructions executed in the computer or other programmable devices provide the steps for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 one flow or multiple flows and / or the functions specified in the block

[0153] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it, although the above embodiments of the present application have been described in detail, those skilled in the art should understand: the specific embodiments of the present application can be modified or replaced by the same, without departing from the spirit and scope of the present application, any modification or equivalent replacement, which should be covered within the scope of protection of the present application.

Claims

1. A method of predicting a discharge voltage of an air gap inside a converter valve module, characterized by The method comprises: Obtaining a plurality of groups of internal air gap discharge voltage data of converter valve modules and performing normalization processing; wherein each group of internal air gap discharge voltage data of the converter valve module comprises: discharge voltage, key structure parameters and meteorological characteristics; Divide the super parameter training set for super parameter tuning from the plurality of groups of internal air gap discharge voltage data of the converter valve module, and perform super parameter tuning of the LightGBM model based on the super parameter training set to determine the optimal super parameter; Based on the optimal super parameter, a LightGBM model is established, and the internal parameters of the LightGBM model are determined based on the model training set divided from the plurality of groups of internal air gap discharge voltage data of the converter valve module, so as to determine the discharge voltage prediction model; The key structure parameters and meteorological characteristics of the converter valve module of the demand structure are input into the discharge voltage prediction model to obtain the discharge voltage prediction value of the gap of the converter valve module of the demand structure.

2. The method of claim 1, wherein, The key structure parameters include: gap structure and gap distance; and the meteorological characteristics include: dry temperature, wet temperature, relative humidity and air pressure.

3. The method of claim 1, wherein, The method uses the Min-Max normalization method to normalize each group of internal air gap discharge voltage data of the converter valve module.

4. The method of claim 1, wherein, Divide the super parameter training set for super parameter tuning from the plurality of groups of internal air gap discharge voltage data of the converter valve module, and perform super parameter tuning of the LightGBM model based on the super parameter training set to determine the optimal super parameter, comprising: The plurality of groups of internal air gap discharge voltage data of the converter valve module are divided into a model training set and a super parameter training set by using a k-fold cross-validation method; Configure the super parameters of LightGBM, select gradient decision boosting tree LGBT as the enhancement method, select learning rate lr, enhancement integration number n_boost, leaf number n_leafs of decision tree growth, maximum depth depth, and minimum segmentation gain min_gain as the super parameters to be tuned; According to the requirements of the model prediction task, the mean square error (MSE) of the LightGBM model output result is taken as the optimization objective function: where y i is the actual value, is the predicted value, n is the number of samples, and the objective function is minimized. The hyperparameter set is tested, the appropriate hyperparameter grid point is selected, and the solution is iterated to find the hyperparameter with the smallest MSE as the optimal hyperparameter of LightGBM.

5. The method of claim 1, wherein, Based on the model training set divided from the plurality of groups of internal air gap discharge voltage data of the converter valve module, the internal parameters of the LightGBM model are determined to determine the discharge voltage prediction model, comprising: The plurality of groups of internal air gap discharge voltage data of the converter valve module are taken as the model training set, and the key structure parameters and meteorological characteristics are selected as the feature input; Take the mean square error MSE as the optimization objective function, set the first-order and second-order regularization parameters λ1 and λ2 to 0.1, and perform training to generate n_boost decision trees, train the internal parameters of each decision tree, and subsequently fit the residual of each tree to the previous tree, and obtain the discharge voltage prediction model based on the weighted combination of multiple trees; Wherein, the splitting gain Gain of the decision tree during training is determined by the following formula: where G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H L is the second order gradient of the left tree, H R is the second order gradient of the right tree; The training process is represented as: wherein, is the prediction value for the t-th round, a i is the weight of the i-th tree, I(x∈R i is an indicator function that takes the value 1 when x belongs to the region R i of the i-th tree, and 0 otherwise.

6. The method of claim 1, wherein, The method further comprises: The discharge voltage prediction value of the gap between the converter valve modules of the demand structure is compared with the discharge voltage design demand value to determine an insulation margin prediction value.

7. A system for predicting discharge voltage of an internal air gap of a converter valve module, characterized by The system comprises: A data processing unit is configured to acquire and normalize a plurality of sets of discharge voltage data of internal air gaps of converter valve modules; each set of discharge voltage data of internal air gaps of converter valve modules comprises a discharge voltage, key structural parameters, and meteorological characteristics; An optimal hyperparameter determination unit is configured to divide a hyperparameter training set for hyperparameter tuning from the plurality of sets of discharge voltage data of internal air gaps of converter valve modules, and perform LightGBM model hyperparameter tuning based on the hyperparameter training set to determine optimal hyperparameters; A model determination unit is configured to establish a LightGBM model based on the optimal hyperparameters, train the model based on a model training set divided from the plurality of sets of discharge voltage data of internal air gaps of converter valve modules, determine internal parameters of the LightGBM model, and thus determine a discharge voltage prediction model; A voltage prediction unit is configured to input key structural parameters and meteorological characteristics of a demand structure of a converter valve module into the discharge voltage prediction model to acquire a discharge voltage prediction value of a gap between converter valve modules of the demand structure.

8. The system of claim 7, wherein, The key structural parameters comprise gap structure and gap distance, and the meteorological characteristics comprise dry temperature, wet temperature, relative humidity, and air pressure.

9. The system of claim 7, wherein, The data processing unit normalizes each set of discharge voltage data of internal air gaps of converter valve modules using a Min-Max normalization system.

10. The system of claim 7, wherein, The optimal hyperparameter determination unit divides a hyperparameter training set for hyperparameter tuning from the plurality of sets of discharge voltage data of internal air gaps of converter valve modules, and performs LightGBM model hyperparameter tuning based on the hyperparameter training set to determine optimal hyperparameters, including: The plurality of sets of discharge voltage data of internal air gaps of converter valve modules are divided into a model training set and a hyperparameter training set using a k-fold cross-validation system; The hyperparameters of LightGBM are configured, the gradient decision boosting tree (LGBT) is selected as an enhanced system, the learning rate (lr), the number of enhanced integrations (n_boost), the number of leaves of decision tree growth (n_leafs), the maximum depth (depth), and the minimum segmentation gain (min_gain) are selected as hyperparameters to be tuned; According to the requirements of the model prediction task, the mean square error (MSE) of the output results of the LightGBM model is used as an optimization objective function: where y i is the actual value, is the predicted value, n is the number of samples, and the objective function is minimized. The hyperparameter set is tested, the appropriate hyperparameter grid points are selected, and the minimum MSE hyperparameters are found as the optimal hyperparameters of LightGBM by traversing the solution.

11. The system of claim 7, wherein, The model determination unit trains the LightGBM model based on the model training set divided from the plurality of sets of discharge voltage data of internal air gaps of converter valve modules to determine internal parameters of the LightGBM model, and thus determines a discharge voltage prediction model, including: The plurality of sets of discharge voltage data of internal air gaps of converter valve modules are used as a model training set, and the key structural parameters and meteorological characteristics are selected as feature inputs; With mean square error MSE as an optimization objective function, first-order and second-order regularization parameters λ1 and λ2 are set as 0.1, and training is performed to iteratively generate n_boost decision trees, internal parameters of each decision tree are obtained through training, each subsequent tree is fitted to the residual of the previous tree, and a discharge voltage prediction model is obtained based on weighted combination of multiple trees; Wherein, the split gain Gain of the decision tree during training is determined by the following formula: where G L is the negative gradient of the left tree, G R is the negative gradient of the right tree, H L is the second order gradient of the left tree, H R is the second order gradient of the right tree; The training process is represented as: wherein, is the prediction value for the t-th round, a i is the weight of the i-th tree, I(x∈R i is an indicator function that takes the value 1 when x belongs to the region R i of the i-th tree, and 0 otherwise.

12. The system of claim 7, wherein, The system further comprises: The insulation margin determination unit is configured to compare the discharge voltage prediction value of the commutation valve module gap of the demand structure with the discharge voltage design demand value to determine an insulation margin prediction value.

Images