A CNN-based diagnosis method for AC transmission line short-circuit faults in AC / DC hybrid power grids

Abstract

The present invention discloses a CNN-based method for diagnosing short-circuit faults in AC transmission lines of an AC / DC hybrid power grid, comprising the following steps: S1. Collecting short-circuit fault recording data on the AC transmission lines of the AC / DC hybrid power grid under different operating conditions, and aggregating the data into a fault sample set after data processing; S2. Establishing a CNN model, using some sample data in the fault sample set to train the CNN model, and the remaining data to test the generalization of the model; S3. Using the trained CNN model to diagnose the fault data and determine the corresponding fault type. By adopting the above technical solution, a CNN model is used to diagnose various short-circuit faults in the AC transmission lines of the AC / DC hybrid power grid and determine the fault type. This method can complete fault diagnosis in a relatively short time and has extremely high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a CNN-based method for diagnosing short-circuit faults in AC transmission lines of an AC / DC hybrid power grid. Background Art

[0002] Before the development of HVDC technology, AC (alternating current) was well-known for its economical and efficient power transmission, playing a crucial role in long-distance transmission. For a long time, AC was widely used in commercial and industrial power supply, dominating virtually the entire power system. However, with the development of large-capacity, long-distance HVDC technology, an increasing number of HVDC transmission lines have been put into operation. Today's power system is no longer a traditional AC system, but a more complex AC / DC hybrid system.

[0003] On the other hand, with the construction of smart grids, more and more data acquisition equipment and communication equipment are used in power grid line operation and maintenance. The Supervisory Control and Data Acquisition (SCADA) and Wide Area Measurement System (WAMS) composed of them have also made important contributions to the safe operation and fault clearing of power systems.

[0004] The massive amounts of monitoring data collected by these systems provide a solid data foundation for the application of deep learning in power system line fault diagnosis and fault location. Previous research has primarily used simple fully connected neural networks and information entropy methods to diagnose faults in pure AC transmission lines. Research on line fault diagnosis in AC / DC hybrid power grids is less mature, and insufficient attention has been paid to fault diagnosis speed. The proposed method still has significant room for improvement in fault diagnosis speed. However, this situation poses a hidden danger to the safe operation and maintenance of AC / DC hybrid power grids. For example, if the most common single-phase ground fault is not promptly detected and cleared by relevant technicians, it may develop into a more detrimental fault such as a two-phase short circuit over time. In severe cases, it may even paralyze the power grid, endangering the normal production activities of factories, enterprises, and people's daily lives. Therefore, it is imperative to provide a powerful tool for fault diagnosis on AC / DC power transmission lines to address this problem. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention proposes a CNN-based diagnosis method for AC transmission line short-circuit faults in AC / DC hybrid power grids. The method collects relevant data before and after the fault on the three-phase transmission line of the AC / DC hybrid power grid, samples them and constructs a fault data set. After normalizing the voltage data and current data respectively, the data is used to train a CNN model. The trained model has good fault diagnosis speed and accuracy.

[0006] In order to solve the above technical problems, the technical solution of the present invention is:

[0007] A CNN-based method for diagnosing short-circuit faults in AC transmission lines of an AC / DC hybrid power grid comprises the following steps:

[0008] S1. Collect short-circuit fault recording data on AC transmission lines of AC / DC hybrid power grids under different working conditions, and summarize them into fault sample sets after data processing.

[0009] S1-1, collecting fault signals from before to after the short-circuit fault occurs, sampling them at a predetermined sampling frequency to obtain fault data;

[0010] S1-2. Linearly normalize the voltage and current data in the fault data. The specific formula is as follows:

[0011]

[0012]

[0013] Among them U min and I min are the minimum values ​​of voltage and current data, U max and I max are the maximum values ​​of voltage and current data, U i and I i are the voltage and current values ​​that have not been normalized, and It's U i and I i Normalized value;

[0014] S1-3. The normalized fault data are labeled according to the corresponding short-circuit fault type and summarized to obtain a fault sample set;

[0015] S1-4. Randomize the order of each sample in the obtained fault sample set according to certain rules to improve the training effect of the CNN model;

[0016] S2. Build a CNN model, use part of the sample data in the fault sample set to train the CNN model, and use the remaining data to test the generalization of the model.

[0017] S3. Use the trained CNN model to diagnose the fault data and determine the corresponding fault type.

[0018] Preferably, in step S1-1, the fault signal is a three-phase voltage, a three-phase current, a zero-sequence voltage and a zero-sequence current signal on the corresponding AC transmission line.

[0019] Preferably, the fault types labeled in step S1-3 include 3 types of single-phase ground short circuit faults, 3 types of interphase short circuit faults, 3 types of double-phase ground short circuit faults, and 1 type of three-phase short circuit fault.

[0020] Preferably, the step S1 also includes step S1-4, after extracting each type of fault from the fault sample set containing the label, the fault subsets with only a single fault are randomly shuffled, and then the single-type fault data set after the shuffled order is divided into a single-type training set, a single-type cross-validation set and a single-type test set according to a certain ratio, and then the training subsets of the 10 fault types are grouped into a complete training set, and the same operation is performed on the cross-validation set and the test set to obtain a complete cross-validation set and a complete test set.

[0021] Preferably, the CNN model includes 2 convolutional layers, 2 fully connected layers and 1 output layer.

[0022] Preferably, the first layer of the CNN model is a convolutional layer with 8 convolution kernels, a convolution kernel size of 2, a step size of 1, and ReLU activation function and L2 regularization are used therein, and batch normalization and maximum pooling layer are also set after the convolution layer, wherein the convolution kernel size of the maximum pooling layer is 2, the step size is 1, and there is no padding after pooling; the second layer of the CNN model is a convolutional layer with 19 convolution kernels, a convolution kernel size of 3, a step size of 1, and ReLU activation function and L2 regularization are used therein, and batch normalization and maximum pooling layer are also set after the convolution layer, wherein the convolution kernel size of the maximum pooling layer is 3, the step size is 2, and there is no padding after pooling.

[0023] Preferably, the third layer of the CNN model is a fully connected layer with 256 neurons, in which the ReLU activation function and L2 regularization are used; the fourth layer of the CNN model is a fully connected layer with 128 neurons, in which the ReLU activation function and L2 regularization are used.

[0024] Preferably, the fifth layer of the CNN model is the output layer, which has 10 neurons, and the 10 neurons respectively correspond to the 10 short-circuit fault types to be diagnosed by the CNN model. This layer uses Softmax as the activation function and uses L2 regularization.

[0025] Preferably, the CNN model uses an Adam optimizer and adopts a cross entropy function as a loss function.

[0026] Preferably, in step S2, the complete training set and the complete cross-validation set processed in step S1 are used to train the constructed CNN model, and multiple rounds of iterations are performed; after the CNN model completes the iterative training, the complete test set is used to test the generalization ability of the model.

[0027] The present invention has the following characteristics and beneficial effects:

[0028] The above technical solution utilizes a CNN model to diagnose and identify various short-circuit faults in AC transmission lines of a hybrid AC / DC power grid. This method not only completes fault diagnosis in a relatively short time, but also demonstrates extremely high accuracy. Furthermore, it fully utilizes data collected by the power grid monitoring system and provides powerful assistance to line operation and maintenance technicians, enabling rapid fault diagnosis and troubleshooting, significantly contributing to the safe operation of the AC / DC hybrid power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 The flowchart of the CNN-based method for diagnosing short-circuit faults in AC transmission lines of an AC / DC hybrid power grid according to the embodiment of the present invention is shown.

[0031] Figure 2 This is a model diagram of an AC / DC hybrid power grid in an embodiment of the present invention.

[0032] Figure 3 This is a graph showing the accuracy and loss function value of the CNN model for the training set and cross-validation set in the embodiment of the present invention.

[0033] Figure 4 This is the confusion matrix obtained by diagnosing the test set using the CNN model in the embodiment of the present invention. DETAILED DESCRIPTION

[0034] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0035] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like 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 invention 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 limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0036] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0037] The present invention provides a method for diagnosing short-circuit faults in AC transmission lines of an AC / DC hybrid power grid based on CNN. Figure 1 As shown, the following steps are included:

[0038] S1. Collect short-circuit fault recording data on AC transmission lines of AC / DC hybrid power grids under different working conditions, and summarize them into fault sample sets after data processing.

[0039] S1-1. Fault signals are collected from a period of time from before to after the short-circuit fault occurs, and are sampled at a predetermined sampling frequency to obtain fault data. Specifically, the fault signals are three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current signals on the corresponding AC transmission line.

[0040] Specifically, the fault signal acquisition process involves collecting three-phase voltage and current data from 0.02 seconds before the short-circuit fault occurs to 0.28 seconds after the fault occurs. Zero-sequence voltage and current data are obtained by summing the three-phase voltage and current data, respectively. The sampling frequency is set to 1 kHz.

[0041] S1-2. Linearly normalize the voltage and current data in the fault data. The specific formula is as follows:

[0042]

[0043]

[0044] Among them U min and I min are the minimum values ​​of voltage and current data, U max and I max are the maximum values ​​of voltage and current data, U i and I i are the voltage and current values ​​that have not been normalized, and It's U i and I i Normalized value;

[0045] S1-3. The normalized fault data are labeled according to the corresponding short-circuit fault type and summarized to obtain a fault sample set. The labeled fault types include 3 single-phase ground short-circuit faults, 3 phase-to-phase short-circuit faults, 3 double-phase ground short-circuit faults, and 1 three-phase short-circuit fault.

[0046] S1-4. After extracting each type of fault from the fault sample set containing labels, randomly shuffle the fault subsets with only a single fault. Then, divide the shuffled single-type fault data set into a single-type training set, a single-type cross-validation set, and a single-type test set according to a certain ratio. Then, group the training subsets of the 10 fault types into a complete training set. Perform the same operation on the cross-validation set and the test set to obtain the complete cross-validation set and the complete test set.

[0047] The ratio of the single-type training set, single-type cross-validation set, and single-type test set is 7:2:1. In the final processed sample dataset, the number of samples in the complete training set, complete validation set, and complete test set is 18120, 5200, and 2600, respectively, with an actual ratio of approximately 7:2:1.

[0048] S2. Establish a CNN model, use the complete training set and complete cross-validation set processed in step S1 to train the established CNN model, and perform multiple rounds of iterations; after the CNN model completes the iterative training, use the complete test set to test the generalization ability of the model.

[0049] Specifically, the CNN model includes 2 convolutional layers, 2 fully connected layers and 1 output layer.

[0050] Furthermore, the first layer of the CNN model is a convolutional layer with 8 convolution kernels, a convolution kernel size of 2, a stride of 1, and a ReLU activation function and L2 regularization. Batch normalization and a maximum pooling layer are also set after the convolution layer, wherein the convolution kernel size of the maximum pooling layer is 2, the stride of 1, and there is no padding after pooling. The second layer of the CNN model is a convolutional layer with 19 convolution kernels, a convolution kernel size of 3, a stride of 1, and a ReLU activation function and L2 regularization are used therein. Batch normalization and a maximum pooling layer are also set after the convolution layer, wherein the convolution kernel size of the maximum pooling layer is 3, the stride of 2, and there is no padding after pooling. The third layer of the CNN model is a fully connected layer with 256 neurons, and a ReLU activation function and L2 regularization are used therein. The fourth layer of the CNN model is a fully connected layer with 128 neurons, and a ReLU activation function and L2 regularization are used therein. The fifth layer of the CNN model is the output layer, which has 10 neurons corresponding to the 10 short-circuit fault types the CNN model is designed to diagnose. This layer uses Softmax as the activation function and L2 regularization. The CNN model uses the Adam optimizer and the cross-entropy function as the loss function.

[0051] It should be noted that ReLU stands for "Rectified Linear Unit." The ReLU activation function is a commonly used neural activation function. The ReLU function is actually a piecewise linear function that converts all negative values ​​to 0 while leaving positive values ​​unchanged. This operation is known as unilateral inhibition. Therefore, in this example, the ReLU activation function is not described in detail.

[0052] Similarly, L2 regularization is a common regularization method in deep learning networks. Therefore, in this embodiment, the ReLU activation function is not described and illustrated in detail.

[0053] In addition, the Softmax activation function is also a conventional technical means. In multi-classification problems, we usually use the Softmax function as the activation function of the network output layer. The Softmax function can normalize the output values ​​and convert all output values ​​into probabilities. The sum of all probability values ​​is equal to 1. The formula of the Softmax function is

[0054]

[0055] Similarly, the Adam optimizer is also common knowledge in this field.

[0056] S3. Use the trained CNN model to diagnose the fault data and determine the corresponding fault type.

[0057] In order to make the content of the present invention clearer and more intuitive, the invention content is further elaborated in conjunction with specific embodiments. The AC / DC hybrid power grid model in the non-parallel operation mode built in Matlab / Simulink is used to replace the real AC / DC hybrid power grid. The specific circuit connection of the model is as follows: Figure 2 shown.

[0058] In addition, short-circuit fault data is generated by simulating a three-phase fault module on a pure AC transmission line in the model. Three-phase voltage and current data are then collected from 0.02 seconds before the short-circuit fault to 0.28 seconds after it occurs. Zero-sequence voltage and current data are obtained by summing the three-phase voltage and current data, respectively. The sampling frequency is set to 1 kHz, resulting in a data size of (300, 8) for a single fault sample. To convert the data into a three-dimensional matrix suitable for CNN processing, the first half of the data is retained, with a size of (256, 8), and then converted into a three-dimensional matrix of size (16, 16, 8).

[0059] In order to make the faults generated by the model closer to real short-circuit faults, a variety of short-circuit fault scenarios are set up. The specific parameters are shown in Table 1. By traversing the parameters in the table, 25,920 short-circuit fault samples under different working conditions can be generated, and then they are divided into training sets, cross-validation sets, and test sets in a ratio of 7:2:1. However, in order to ensure that each fault type has the same proportion in these three data sets, it is necessary to store the complete fault data into 10 sub-datasets according to the fault type. After each sub-dataset is shuffled in a ratio of 7:2:1, the training set, validation set, and test set of each sub-dataset are then converted into training sets, cross-validation sets, and test sets containing 10 fault types. Due to the integer nature of the number of samples, the final number of samples in the training set, validation set, and test set are 18,120, 5,200, and 2,600, respectively, and the actual ratio between them is approximately 7:2:1.

[0060] Table 1: Short circuit fault generation related parameters

[0061]

[0062] like Figure 3 The figure shows the accuracy change curve and loss function value change curve of the CNN model after 50 epochs of training for the training set and cross-validation set in this specific embodiment. During multiple iterative training, the data is passed to the CNN model in batches for processing, and the batch size is 64. The final accuracy of the CNN model in this specific embodiment for fault diagnosis on the training set reaches 99.92%, and the accuracy on the cross-validation set is 99.61%. Figure 4The figure shows the confusion matrix obtained by the trained CNN model for fault diagnosis of the test set. In addition, when performing fault diagnosis on 2600 test set samples, the CNN model in this specific embodiment only takes an average of 22.5ms per diagnosis.

[0063] This specific embodiment proves that the CNN-based diagnosis method for AC transmission line short-circuit faults in AC / DC hybrid power grids proposed in the present invention has both high recognition accuracy and good generalization ability for unknown fault samples, as well as high diagnosis speed. It can better meet the accuracy and real-time requirements of line fault diagnosis in the current smart grid, and also illustrates the practicality of the method proposed in the present invention.

[0064] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions, and variations of these embodiments, including components, without departing from the principles and spirit of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for diagnosing short-circuit faults in AC transmission lines of an AC / DC hybrid power grid based on CNN, characterized in that: The steps include: S1. Collect short-circuit fault recording data on AC transmission lines of AC / DC hybrid power grids under different working conditions, and summarize them into fault sample sets after data processing. S1-1, collecting fault signals from before to after the short-circuit fault occurs, sampling them at a predetermined sampling frequency to obtain fault data; S1-2. Linearly normalize the voltage and current data in the fault data. The specific formula is as follows: Among them U min and I min are the minimum values ​​of voltage and current data, U max and I max are the maximum values ​​of voltage and current data, U i and I i are the voltage and current values ​​that have not been normalized, and It's U i and I i Normalized value; S1-3. The normalized fault data are labeled according to the corresponding short-circuit fault type and summarized to obtain a fault sample set; S1-4. After extracting each type of fault from the labeled fault sample set, randomly shuffle the fault subsets containing only a single fault. Then, divide the shuffled single-type fault data set into a single-type training set, a single-type cross-validation set, and a single-type test set according to a certain ratio. Then, combine the training subsets of the 10 fault types into a complete training set. Perform the same operation on the cross-validation set and the test set to obtain the complete cross-validation set and the complete test set. S2. Build a CNN model, use part of the sample data in the fault sample set to train the CNN model, and use the remaining data to test the generalization of the model. S3. Use the trained CNN model to diagnose the fault data and determine the corresponding fault type.

2. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 1, characterized in that: In step S1 - 1 , the fault signal is the three-phase voltage, three-phase current, zero-sequence voltage and zero-sequence current signals on the corresponding AC transmission line.

3. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 1, characterized in that: The fault types labeled in step S1-3 include three types of single-phase ground short circuit faults, three types of interphase short circuit faults, three types of double-phase ground short circuit faults, and one type of three-phase short circuit fault.

4. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 1, characterized in that: The CNN model includes 2 convolutional layers, 2 fully connected layers and 1 output layer.

5. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 4, characterized in that: The first layer of the CNN model is a convolutional layer with 8 convolution kernels, a convolution kernel size of 2, a step size of 1, and a ReLU activation function and L2 regularization. Batch standardization and a maximum pooling layer are also set after the convolution layer, wherein the convolution kernel size of the maximum pooling layer is 2, the step size is 1, and there is no padding after pooling; the second layer of the CNN model is a convolutional layer with 19 convolution kernels, a convolution kernel size of 3, a step size of 1, and a ReLU activation function and L2 regularization are used therein. Batch standardization and a maximum pooling layer are also set after the convolution layer, wherein the convolution kernel size of the maximum pooling layer is 3, the step size is 2, and there is no padding after pooling.

6. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 4, characterized in that: The third layer of the CNN model is a fully connected layer with 256 neurons, in which the ReLU activation function and L2 regularization are used; the fourth layer of the CNN model is a fully connected layer with 128 neurons, in which the ReLU activation function and L2 regularization are used.

7. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 4, characterized in that: The fifth layer of the CNN model is the output layer, which has 10 neurons. The 10 neurons correspond to the 10 short-circuit fault types to be diagnosed by the CNN model. This layer uses Softmax as the activation function and uses L2 regularization.

8. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 1, characterized in that: The CNN model uses the Adam optimizer and adopts the cross entropy function as the loss function.

9. The method for diagnosing short-circuit faults of AC transmission lines in an AC / DC hybrid power grid based on CNN according to claim 1, characterized in that: In step S2, the complete training set and the complete cross-validation set processed in step S1 are used to train the constructed CNN model, and multiple rounds of iterations are performed; after the CNN model completes the iterative training, the complete test set is used to test the generalization ability of the model.

Images