Method for predicting stray inductance of submodule of high-voltage direct-current circuit breaker based on support vector regression

Abstract

The invention discloses a high-voltage direct-current circuit breaker sub-module stray inductance prediction method based on support vector regression, and the method comprises the following steps: building a sub-module structure model in finite element simulation software, and defining three structural geometric parameters: the thickness of a diversion layer, the length of a busbar and the diameter of a connecting column as the input of the sub-module structure model; the overall stray inductance of the combined high-voltage direct-current circuit breaker sub-module is output by the sub-module structure model; in the data space, a hyperplane is defined, the hyperplane fits the experimental data obtained through training according to the set fault tolerance degree, and meanwhile the interval between the hyperplane and the data points is maximized; carrying out support vector regression on the hyperplane objective function by adopting an S-type kernel function; using cross validation to obtain optimal penalty parameters, fault tolerance and S-type kernel function parameters; new penalty parameters and S-type kernel function parameters are adopted for retraining, and an optimal stray inductance prediction result is obtained. According to the invention, only a small volume data set is needed to complete model establishment.

Description

Technical Field

[0001] The invention belongs to the technical field of direct current circuit breakers and relates to a method for predicting stray inductance of a high-voltage direct current circuit breaker submodule based on support vector regression. Background Art

[0002] High-voltage direct current transmission has attracted more and more attention due to its advantages such as low construction cost, stable and reliable operation, and good functional regulation. However, the rapid current rise rate when the DC grid fails and the lack of a natural zero crossing point limit its development. There is an urgent need to develop key equipment including high-voltage direct current circuit breakers.

[0003] Combined high-voltage DC circuit breaker is a new type of equipment. Various structures and layouts need to be rearranged. Among them, the sub-module, as the core component of the circuit breaker, plays a very critical role in the breaking process. Reasonable layout can greatly improve the working stability of the equipment.

[0004] Rapidly and accurately predicting the stray inductance generated by the model structure is very important for the normal operation of the combined high-voltage DC circuit breaker. The current methods for predicting stray inductance mainly include: 1. Using analytical methods, based on electromagnetic field theory, using formulas to directly calculate stray inductance; 2. Using the finite element method, dividing the structure into many small units, and then solving Maxwell's equations to calculate the electromagnetic field distribution, thereby obtaining the stray inductance; 3. Using the measurement method, through experimental measurement of the circuit response, the value of the stray inductance is inferred. Common methods include vector network analyzer (VNA) measurement.

[0005] However, the above stray inductance measurement methods have the following disadvantages: 1. Low accuracy. The simplified assumptions on geometric shapes will introduce errors. It is only applicable to simple geometric structures. It is difficult to accurately calculate complex three-dimensional structures, and it is usually difficult to consider the influence of the surrounding environment. 2. The calculation speed is slow, requiring professional FEM software and certain professional knowledge, and the model construction takes a long time. 3. Specialized measurement equipment is required, the cost is high, and the measurement process is relatively complicated. It can only measure the stray inductance of specific circuits or devices, and it is difficult to generalize to other designs. Summary of the invention

[0006] In view of the above technical problems, the present invention proposes a method for predicting the stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression, so that the stray inductance generated by the model structure can be quickly and accurately predicted; when the model structure parameters change, the stray inductance extraction method is still applicable; under the premise of meeting actual engineering needs, a solution with the lowest cost and less stray inductance is obtained.

[0007] A method for predicting stray inductance of a high-voltage direct current circuit breaker submodule based on support vector regression is applied to a combined high-voltage direct current circuit breaker submodule, comprising the following steps:

[0008] Step 1: Build a submodule structural model in the finite element simulation software. According to the principle of electromagnetic distribution, the thickness of the guide layer, the length of the busbar and the diameter of the connecting column in the combined high-voltage DC circuit breaker submodule are respectively changed in proportion through the optimization mode in the finite element simulation software to obtain the overall stray inductance under different structural dimensions as experimental data. Define three structural geometric parameters: the thickness of the guide layer, the length of the busbar and the diameter of the connecting column are the input of the submodule structural model, and the overall stray inductance of the combined high-voltage DC circuit breaker submodule is the output of the submodule structural model. The experimental data are divided into a training set and test set ,in and are the guide layer thickness, busbar length and connecting column diameter in the training and test sets, respectively. and They are the equivalent total inductance of the submodule structure model in the training set and the test set after a certain frequency current is passed through it;

[0009] Step 2: In the data space formed by the experimental data described in step 1, a hyperplane is defined, which is based on the fault tolerance Fit the experimental data obtained from training and maximize the interval between the hyperplane and the data points. The hyperplane objective function is described as follows:

[0010] ,

[0011] in, is the weight vector of the hyperplane, is the intercept of the hyperplane, is the penalty parameter, which controls the degree of penalty for the error of the submodule structure model. is a slack variable, indicating the degree of deviation of the data points;

[0012] Step 3: Use the S-type kernel function to perform support vector regression on the hyperplane objective function in step 2;

[0013] Step 4: Use cross validation to get the optimal penalty parameter , fault tolerance And the parameters of the S-type kernel function;

[0014] Step 5: Retrain using new penalty parameters and S-type kernel function parameters to obtain the optimal stray inductance prediction result.

[0015] Beneficial effects brought by the technical solution of the present invention:

[0016] 1. The present invention adopts a programming language modeling method to predict the stray inductance result. Only a small volume data set is needed to complete the model establishment, and the model fitting goodness can reach 98.76%.

[0017] 2. The present invention can combine actual engineering needs and cost requirements to quickly and accurately provide a structural solution with the lowest stray inductance.

[0018] 3. The model in the present invention has three characteristic variables, which greatly reduces the complexity of extracting stray inductance. The model has good compatibility. When the sub-module structure layout changes, it only needs to import a new data set to complete the prediction model of the new structure, and the characteristic variables in the data set can be appropriately increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A model structure diagram of a combined high-voltage DC circuit breaker submodule according to an embodiment of the present invention is shown;

[0020] Figure 2 A flow chart of a method for predicting stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression according to an embodiment of the present invention is shown;

[0021] Figure 3 A comparison diagram of model prediction results according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0022] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0023] According to an embodiment of the present invention, a method for predicting stray inductance of a high-voltage direct current circuit breaker submodule based on support vector regression is provided, which is applied to a combined high-voltage direct current circuit breaker submodule. Figure 1 The model structure diagram of the combined high-voltage DC circuit breaker submodule according to the embodiment of the present invention is shown. The submodule is composed of power devices crimped together, and the side surface adopts a three-layer vertical stacked busbar connection device. Holes are left on the busbar for inserting connecting columns, and support columns are connected around to stabilize the entire device. The current of the entire device flows through the busbar, connecting columns, and devices. The path is relatively complex. The overall layout of the device is fixed, and only the thickness of the guide layer, the length of the busbar, and the diameter of the connecting column can be changed. When the submodule is working normally, current will be passed through. The stray inductance in the structure will affect the current frequency and size under the action of electromagnetic induction. When the thickness of the submodule's guide layer, the length of the busbar, and the diameter of the connecting column change, the stray inductance generated by the entire structure will change accordingly.

[0024] The method for predicting the stray inductance of a high-voltage DC circuit breaker submodule provided by the present invention does not require repeated remodeling to extract the stray inductance, and adopts a support vector regression method (SVR) to predict the structural stray inductance. Figure 2FIG. 4 is a flow chart of a method for predicting stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression according to an embodiment of the present invention. Figure 2 As shown, the method includes:

[0025] Step 1: Build a submodule structure model in the finite element simulation software. According to the principle of electromagnetic distribution, the thickness of the guide layer, the length of the busbar and the diameter of the connecting column in the submodule are proportionally changed through the optimization mode in the finite element simulation software ANSYS Q3D to obtain the overall stray inductance under different structural sizes as experimental data. In one embodiment, the experimental data includes a total of 495 sets of data. Three structural geometric parameters are defined: the thickness of the guide layer, the length of the busbar and the diameter of the connecting column are model inputs, and the overall stray inductance of the module is model output, and the experimental data is randomly divided into a training set and test set , the training set accounts for 70% and the test set accounts for 30%, of which and are the guide layer thickness, busbar length and connecting column diameter in the model training set and test set, respectively. and The equivalent total inductance after a certain frequency current is passed into the model training set and test set respectively, and then all feature variables in the training set and test set are standardized with a mean of 0 and a standard deviation of 1;

[0026] Step 2: In the data space formed by the experimental data described in step 1, a hyperplane is defined, which is based on the fault tolerance ( ) fits the experimental data obtained through training, while maximizing the interval between the hyperplane and the set data points. The hyperplane objective function is described as follows:

[0027] ,

[0028] in, is the weight vector of the hyperplane, is the intercept of the hyperplane, is the penalty parameter, which controls the degree of penalty for the error of the submodule structure model. is a slack variable, indicating the degree of deviation of the data points;

[0029] Step 3: S-type kernel function is used to perform support vector regression on the hyperplane objective function in step 2. The penalty parameters of all hyperplane objective functions use default values, and the preliminary goodness of fit is 83.29%;

[0030] Step 4: Use cross validation to get the optimal penalty parameter , fault tolerance The process includes: resetting the penalty parameters and the kernel function parameters, using the k-fold cross-validation method, randomly dividing the experimental data into 10 subsets, and looping 10 times. Each time, one subset is selected as the test set, and the remaining subsets are used as the training set. The submodule structure model is trained and the performance is evaluated. The optimal penalty parameter is obtained according to the optimal performance. , Fault Tolerance and kernel function parameters; this step can improve the accuracy of the submodule structure model;

[0031] Step 5: Use new penalty parameters The kernel function parameters are retrained to obtain the optimal stray inductance prediction result. The optimal stray inductance prediction evaluation indicators are as follows:

[0032] Root mean square error (RMSE): 0.192861,

[0033] Squared absolute error (MAE): 0.183850,

[0034] R squared (R 2 ): 0.987573,

[0035] The goodness of fit is 98.76%. The prediction accuracy of this model is as follows: Figure 3 shown.

Claims

1. A method for predicting stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression, applied to a combined high-voltage DC circuit breaker submodule, characterized in that: The following steps are involved: Step 1: Build a submodule structural model in the finite element simulation software. According to the principle of electromagnetic distribution, the thickness of the guide layer, the length of the busbar and the diameter of the connecting column in the combined high-voltage DC circuit breaker submodule are respectively changed in proportion through the optimization mode in the finite element simulation software to obtain the overall stray inductance under different structural dimensions as experimental data. Define three structural geometric parameters: the thickness of the guide layer, the length of the busbar and the diameter of the connecting column are the input of the submodule structural model, and the overall stray inductance of the combined high-voltage DC circuit breaker submodule is the output of the submodule structural model. The experimental data are divided into a training set and test set ,in and are the guide layer thickness, busbar length and connecting column diameter in the training and test sets, respectively. and They are the equivalent total inductance of the submodule structure model in the training set and the test set after a certain frequency current is passed through it; Step 2: In the data space formed by the experimental data described in step 1, a hyperplane is defined, which is based on the fault tolerance Fit the experimental data obtained from training and maximize the interval between the hyperplane and the data points. The hyperplane objective function is described as follows: ， in, is the weight vector of the hyperplane, is the intercept of the hyperplane, is the penalty parameter, which controls the degree of penalty for the error of the submodule structure model. is a slack variable, indicating the degree of deviation of the data points; Step 3: Use the S-type kernel function to perform support vector regression on the hyperplane objective function in step 2; Step 4: Use cross validation to get the optimal penalty parameter , fault tolerance And the parameters of the S-type kernel function; Step 5: Retrain using new penalty parameters and S-type kernel function parameters to obtain the optimal stray inductance prediction result.

2. The method for predicting stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression according to claim 1 is characterized in that: The experimental data in step 1 includes a total of 495 sets of data.

3. The method for predicting stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression according to claim 1 is characterized in that: The training set and test set of step 1 account for 70% and 30% respectively.

4. The method for predicting stray inductance of a high-voltage DC circuit breaker submodule based on support vector regression according to claim 1, characterized in that: Step 4 includes: Reset the values ​​of hyperparameters and kernel function parameters, use k-fold cross validation method, randomly divide the experimental data into 10 subsets, cycle 10 times, select one subset as the test set each time, and the remaining subsets as the training set, train the submodule structure model separately and evaluate the performance, and obtain the optimal penalty parameter based on the optimal performance , Fault Tolerance and kernel function parameters.

Images