Method and system for diagnosing commutation failure of high-voltage direct-current power transmission system

Abstract

The invention relates to a high-voltage direct-current power transmission system commutation failure fault diagnosis method and system, and belongs to the technical field of power systems. According to the method, the decision binary tree in the fault diagnosis model is improved, and the support vector machine is used as a discrimination function in the decision binary tree, so that the discrimination result of the fault diagnosis model on the commutation failure reason type is finer and more accurate; according to the method, the time sequence data for the commutation failure fault diagnosis of the high-voltage direct-current power transmission system is obtained and input into the trained improved fault diagnosis model, the commutation failure reason of the high-voltage direct-current power transmission system is obtained, and the improved fault diagnosis model is relatively accurate in commutation fault diagnosis result; maintenance personnel can be helped to quickly locate the root of the problem, measures are taken in time to prevent fault expansion, and the negative influence of the commutation failure on the stability and reliability of a power grid is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to a high-voltage direct current transmission system commutation failure fault diagnosis method and system, and belongs to the technical field of power systems. BACKGROUND

[0002] With the continuous growth of global power demand, high-voltage direct current transmission technology has become an important part of modern power industry due to its unique advantages in large-capacity, long-distance power transmission and regional power grid interconnection. However, as one of the frequently encountered fault types in the operation process of high-voltage direct current transmission system, commutation failure causes sudden increase of direct current, increase of commutation valve stress, severe fluctuation of direct current power, and even system lockout and other chain reactions, which seriously threatens the safety and stability of high-voltage direct current transmission system. Therefore, accurate and rapid identification and effective response to commutation failure faults in high-voltage direct current transmission system have immeasurable value for ensuring stable operation of power grid.

[0003] Existing commutation failure diagnosis methods mainly include arc extinction angle discrimination method, voltage drop method, direct current voltage zero-crossing method and current detection method, but these methods are generally limited by idealization of model assumptions, difficulty in accurately setting key parameters (including critical voltage, arc extinction angle, etc.), and other problems, resulting in insufficient commutation failure diagnosis accuracy or response lag. With the rapid development of sensor technology and the improvement of computer data storage capacity, as well as the richness and quantity of power system monitoring data, signal time-frequency analysis methods and deep learning technology have also been applied to the fault diagnosis of commutation failure. However, signal time-frequency analysis methods and deep learning technology still cannot fully meet the needs of model complexity, computing resources and engineering implementation convenience in actual operation process. Considering the limited computing resources at the station end, how to optimize and improve the model analysis accuracy and reduce the model complexity is also the focus of future research. In addition, with the continuous advancement of industrial intelligence of Internet of Things, although the operation data of power system can be effectively monitored and recorded, it is still a difficult problem to be solved to collect enough commutation failure samples in a short period of time and support scientific and effective diagnosis and analysis of commutation failure. SUMMARY

[0004] The purpose of the present application is to provide a high-voltage direct current transmission system commutation failure fault diagnosis method and system to solve the problem that the existing commutation failure fault diagnosis model cannot fully meet the accuracy requirements of commutation failure diagnosis in the actual operation process of high-voltage direct current transmission system.

[0005] To achieve the above-mentioned purpose, the scheme of the present application includes:

[0006] The high-voltage direct current transmission system commutation failure fault diagnosis method of the present application includes:

[0007] When a commutation failure fault of a high-voltage direct current transmission system occurs, time sequence data for diagnosing the commutation failure fault of the high-voltage direct current transmission system is obtained, and the time sequence data is input to a trained fault diagnosis model to obtain a commutation failure reason of the high-voltage direct current transmission system.

[0008] The fault diagnosis model is a model obtained by improving a decision binary tree, and the improvement includes using a support vector machine for binary classification of a discriminant function in the decision binary tree.

[0009] Further, the time sequence data for diagnosing the commutation failure fault includes inverter-side bus voltage time sequence data and direct current time sequence data.

[0010] Further, the training data used for training the fault diagnosis model is training data obtained by expanding original training data using a generative adversarial network.

[0011] Further, the generative adversarial network is a WGAN, and the discriminator and the generator of the WGAN are both four-layer structures, including three convolutional layers and one fully connected layer connected in sequence.

[0012] Further, the activation function in the first two convolutional layers of the discriminator is a Relu activation function, and the activation function in the last convolutional layer of the discriminator is a Tanh activation function; the activation function in the three convolutional layers of the generator is a Relu activation function.

[0013] Further, the time sequence data input to the trained fault diagnosis model is time sequence data obtained by completing original collected time sequence data, and the completion method is: for a non-continuous null value in the original collected time sequence data, the average value of the data before and after the non-continuous null value is calculated, and the obtained average value is used to complete the non-continuous null value; for a continuous null value in the original collected time sequence data, the average value of the data at the moment before the continuous null value starts and the moment after the continuous null value ends is calculated, and the obtained average value is used to complete the continuous null value.

[0014] Further, the commutation failure reason of the high-voltage direct current transmission system includes a single-phase grounding fault, a two-phase short-circuit fault, a two-phase grounding fault, a three-phase short-circuit fault, a direct current line fault, and a normal state.

[0015] Further, the decision binary tree uses a CART decision classifier.

[0016] Further, the kernel function in the support vector machine uses a Gaussian radial basis kernel function.

[0017] The high-voltage direct current transmission system commutation failure fault diagnosis system of the application comprises a processor, characterized in that the processor is used to execute a computer program to realize the high-voltage direct current transmission system commutation failure fault diagnosis method.

[0018] The high-voltage direct current power transmission system commutation failure fault diagnosis system of the application comprises a processor, and the processor is used to execute a computer program to realize the high-voltage direct current power transmission system commutation failure fault diagnosis method.

[0019] The application has the following beneficial effects:

[0020] The high-voltage direct current power transmission system commutation failure fault diagnosis method and system improve the decision binary tree in the fault diagnosis model, and the specific improvement is that the support vector machine is used as a discriminant function in the decision binary tree, so that the discrimination result of the fault diagnosis model on the commutation failure reason type is more accurate and accurate. Further, the time series data for high-voltage direct current power transmission system commutation failure fault diagnosis is obtained, and is input into the improved fault diagnosis model trained, so that the commutation failure reason of the high-voltage direct current power transmission system can be accurately obtained. The improved fault diagnosis model is more accurate in the result of commutation fault diagnosis, can help maintenance personnel to quickly locate the root cause of the problem, and timely take measures to prevent the fault from expanding, and reduce the negative influence of commutation failure fault on the stability and reliability of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a schematic diagram of the high-voltage direct current power transmission system commutation failure fault diagnosis method of the application;

[0022] Figure 2 is a structure diagram of the DT-SVM model in the high-voltage direct current power transmission system commutation failure fault diagnosis method of the application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical scheme and advantages of the application more clear and clear, the application will be clearly and completely described in detail below in combination with the drawings.

[0024] The idea of the application is to improve the decision binary tree in the fault diagnosis model, and the specific improvement is that the support vector machine is used as a discriminant function in the decision binary tree, so that the discrimination result of the fault diagnosis model on the commutation failure reason type is more accurate and accurate; the improved fault diagnosis model is used to analyze and judge the time series data for high-voltage direct current power transmission system commutation failure fault diagnosis, and more accurate commutation failure reasons are obtained.

[0025] An embodiment of a high-voltage direct current power transmission system commutation failure fault diagnosis method:

[0026] During the operation of the high-voltage direct current power transmission system, commutation failure usually causes a large decrease in the inverter side bus voltage, a rapid increase in the direct current, and a significant decrease in the arc extinction angle, so the commutation failure fault can be diagnosed by the change trend of the arc extinction angle, the inverter side bus voltage and the direct current.

[0027] However, different HVDC systems may have different structures and parameter settings, and the calculation results and judgment criteria of the arc extinction angle are different, and it is still not possible to solve the calculation results and judgment criteria of the arc extinction angle that are universally applicable to all HVDC systems, so it is easy to make mistakes based on the arc extinction angle for commutation failure fault. The parameters such as the inverter side bus voltage and the DC current do not depend on the specific parameter configuration of the HVDC system, so they have more extensive applicability and universality, and can be applied to various types and scales of HVDC systems. Therefore, the commutation failure fault diagnosis method of the HVDC system in the embodiment uses the inverter side bus voltage and the DC current to diagnose and analyze the commutation failure fault.

[0028] By analyzing the existing sample data of commutation failure fault, the fault types that cause commutation failure are divided into single-phase ground fault, two-phase short-circuit fault, two-phase ground fault and three-phase short-circuit fault. Since the state of the HVDC system during DC line fault is very similar to the transient process of the inverter side bus voltage and the DC current during commutation failure, in order to avoid the action of the relay protection device, the DC line fault and the commutation failure fault are distinguished and identified in the embodiment, and the DC line fault is taken as a fault type that causes commutation failure for diagnosis and analysis.

[0029] In addition, the normal state is added as a fault type for commutation failure diagnosis during the diagnosis and analysis of commutation failure fault, so the commutation failure fault types can be divided into six types, including single-phase ground fault, two-phase short-circuit fault, two-phase ground fault, three-phase short-circuit fault, DC line fault and normal state, and the normal state is used to exclude commutation failure fault.

[0030] As shown in Figure 1 The commutation failure fault diagnosis method of the HVDC system of the present application comprises:

[0031] 1) Sample data preprocessing.

[0032] Obtain the data set for training and testing the fault diagnosis model, which specifically includes time series data for HVDC system commutation failure fault diagnosis and the corresponding fault type.

[0033] In the embodiment, the time series data for HVDC system commutation failure fault diagnosis is selected as the time series data of the inverter side bus voltage and the DC current.

[0034] In actual scenarios, due to measurement errors, sensor abnormalities and other reasons, the collected time series data will inevitably have null values; to avoid the impact of null values on subsequent calculations, the null values can be completed using the mean value method of the data before and after the data, and for continuous null values, the mean value of the moment before the missing sequence starts and the moment after it ends can be taken to complete the filling.

[0035] At the same time, in order to improve the convergence speed and the authenticity of the generated data when the subsequent generative adversarial network uses these data for calculation, the data of the two electrical parameters of the inverter side bus voltage and the direct current can be normalized, and in the embodiment, the Min-Max method, i.e. formula (1), is used for normalization:

[0036]

[0037] Where x is the sample data to be normalized; x * is the normalized value; x min , x max are the minimum and maximum values in the data before normalization, respectively.

[0038] 2) Expand the sample data based on the generative adversarial network model.

[0039] A large amount of sample data is needed as a basis for fault judgment in the commutation failure fault diagnosis process to ensure the correctness of the judgment, and it is difficult to collect enough commutation failure fault sample data from the high-voltage direct-current power transmission system in a short period of time.

[0040] To solve the above problems, the generative adversarial network is used in the embodiment to expand the sample data for fault judgment.

[0041] After preprocessing the sample data, the generative adversarial network is used to expand the sample data, random noise is added to the original sample data, and the generator in the generative adversarial network tries to generate data consistent with the distribution of the original sample data extracted from the actual operation process of the system, in order to "deceive" the discriminator in the generative adversarial network; while the discriminator in the generative adversarial network tries to accurately distinguish the original sample data from the sample data generated by the generator in the generative adversarial network, after the continuous game between the generator in the generative adversarial network and the discriminator in the generative adversarial network and reaching Nash equilibrium, the generated sample data can well fit the data generated in the actual working process, and the sample data set is continuously expanded, so as to have enough sample data for feature extraction and identification of various faults.

[0042] As a preferred embodiment, the sample data expansion in the embodiment does not use the traditional generative adversarial network, but uses an improved adversarial generative network, namely the Wasserstein generative adversarial network. Compared with the traditional generative adversarial network, the Wasserstein generative adversarial network introduces the Wasserstein distance (also known as "bulldozer distance") to replace the JS divergence in the traditional adversarial generative network, solves the problems of unstable training and mode collapse of the traditional adversarial generative network, and significantly improves the quality and diversity of the expanded generated samples.

[0043] In the embodiment, the discriminator and the generator in the Wasserstein generative adversarial network are both set to a four-layer structure, wherein the generator is one fully connected layer and three convolutional layers, the number of convolutional kernels of each layer of the convolutional layers is 32, 64 and 128 in turn, the activation function is Tanh function for the last layer and Relu function for the other layers, and the layer size is 2x2, 2x2 and 1x1; the discriminator is three convolutional layers and one fully connected layer, the activation function is all Relu function, the number of convolutional kernels of each layer of the convolutional layers is 128, 64 and 32 in turn, the layer size is 1x1, 2x2 and 2x2, and the activation function of the convolutional layer is Relu.

[0044] As a preferred embodiment, in the Wasserstein generative adversarial network, the learning rate is set to 0.0001, the epoch is 3000, and the Adam optimizer is used.

[0045] 3) Construct a fault judgment model.

[0046] In the embodiment, the DT-SVM decision model is used as the fault judgment model, and the DT-SVM decision model is a hybrid machine learning model combining the decision tree DT and the support vector machine SVM.

[0047] In the embodiment, the DT-SVM decision model is constructed for the commutation failure fault diagnosis by combining the characteristics of the decision tree DT and the support vector machine SVM model. The core of the model is to construct a suitable decision binary tree. Since the discriminant function of all classifiers does not need to be calculated in the classification process, the DT-SVM decision model can be constructed according to the principle from easy to difficult. First, it is judged whether the sample data is a fault sample, and then the fault samples are classified step by step. The structure of the DT-SVM decision model is as shown in Figure 2

[0048] ​Firstly, the decision tree DT is used to preliminarily divide the sample data set, and the sample data set is divided into multiple subsets by selecting the best feature and the division point, each subset corresponding to a node or a leaf of the decision tree. In this process, the decision tree DT obtains the criterion of the normal state by identifying the sample data in the normal state.

[0049] The decision tree DT can help identify the most important features for the classification task and provide more targeted data subsets for subsequent support vector machine classification.

[0050] At each node or leaf of the decision tree DT, a support vector machine SVM classifier can be trained. These support vector machine SVM classifiers are optimized for specific data subsets and can capture more complex patterns in the data. The support vector machine SVM classifier divides the remaining sample data into subsets according to the data characteristics corresponding to each fault, obtaining more accurate criteria for each fault, and obtaining a support vector machine corresponding to each fault, including a support vector machine SVM1 corresponding to a DC line fault, a support vector machine SVM2 corresponding to a single-phase ground fault, a support vector machine SVM3 corresponding to a two-phase open circuit fault, a support vector machine SVM4 corresponding to a two-phase open circuit fault, and a sample data subset other than the sample data corresponding to the normal state, the DC line fault, the single-phase ground fault, the two-phase open circuit fault, and the two-phase open circuit fault.

[0051] The DT-SVM decision model in the embodiment combines the prediction results of multiple support vector machines SVM classifiers, and the DT-SVM decision model can make more accurate classification decisions.

[0052] As a preferred embodiment, the decision tree algorithm in the DT-SVM decision model uses the CART classification algorithm.

[0053] As a preferred embodiment, a Gaussian radial basis function RBF is introduced into the support vector machine SVM to improve the model generalization ability.

[0054] 4) The augmented sample data obtained in step 2) is used as training data to train the fault diagnosis model constructed in step 3), and a trained fault diagnosis model is obtained.

[0055] 5) When a high-voltage direct-current power transmission system commutation failure fault occurs, the time series data of the inverter side bus voltage and direct current and other parameters in the high-voltage direct-current power transmission system are input into the trained fault diagnosis model in step 4) to obtain the commutation failure type.

[0056] An embodiment of a high-voltage direct-current power transmission system commutation failure fault diagnosis system:

[0057] A high-voltage direct current power transmission system commutation failure fault diagnosis system comprises a processor, and the processor executes a computer program to implement a high-voltage direct current power transmission system commutation failure fault diagnosis method in an embodiment of the high-voltage direct current power transmission system commutation failure fault diagnosis method.

[0058] In conclusion, the high-voltage direct current power transmission system commutation failure fault diagnosis method and system of the present application use the improved generative adversarial network to expand the sample data to solve the problem of insufficient sample quantity, and then construct a DT-SVM decision model as a fault diagnosis model to analyze the commutation failure fault; the decision binary tree in the DT-SVM decision model is improved, and the support vector machine is used as a discriminant function in the decision binary tree, so that the discrimination result of the fault diagnosis model on the reason type of the commutation failure is more accurate and accurate, which can effectively improve the accuracy and diagnosis efficiency of the commutation failure fault diagnosis, help maintenance personnel quickly locate the root cause of the problem, take timely measures to prevent the fault from being enlarged, and reduce the negative impact of the commutation failure on the stability and reliability of the high-voltage direct current power transmission system.

Claims

1. A method for diagnosing commutation failure in a high-voltage direct current transmission system, characterized in that: The method includes: When a commutation failure fault occurs in the HVDC transmission system, time series data for diagnosing the commutation failure fault of the HVDC transmission system is obtained and input into a trained fault diagnosis model to obtain the cause of the commutation failure of the HVDC transmission system; Among them, the fault diagnosis model is a model obtained by improving the decision binary tree, and the improvement includes using a support vector machine to perform binary classification on the discriminant function in the decision binary tree.

2. The method for diagnosing commutation failure in a HVDC system according to claim 1, wherein: The commutation failure fault diagnosis time series data includes inverter-side bus voltage time series data and DC current time series data.

3. The method for diagnosing commutation failure in a HVDC system according to claim 1, wherein: The training data used to train the fault diagnosis model is training data obtained by expanding the original acquired training data using a generative adversarial network.

4. The method for diagnosing commutation failure in a HVDC system according to claim 3, wherein: The generative adversarial network is WGAN, and both the discriminator and the generator of WGAN have a four-layer structure, which includes three convolutional layers and one fully connected layer connected in sequence.

5. The method for diagnosing commutation failure in a HVDC system according to claim 4, characterized in that: The activation function in the first two convolutional layers of the discriminator is the Relu activation function, and the activation function in the last convolutional layer of the discriminator is the Tanh activation function; the activation function in the three convolutional layers of the generator is the Relu activation function.

6. The method for diagnosing commutation failure in a HVDC system according to claim 1, wherein: The time series data input into the trained fault diagnosis model is the time series data after the original collected time series data is supplemented, and the supplement method is: for the non-continuous null values ​​in the original collected time series data, the average value of the data before and after the non-continuous null value is calculated, and the obtained average value is used to supplement the non-continuous null value; for the continuous null values ​​in the original collected time series data, the average value of the data before and after the continuous null value is calculated, and the obtained average value is used to supplement the continuous null value.

7. The method for diagnosing commutation failure in a HVDC system according to claim 1, wherein: The causes of commutation failure in HVDC transmission system include single-phase grounding fault, two-phase short circuit fault, two-phase grounding fault, three-phase short circuit fault, DC line fault and normal state.

8. The method for diagnosing commutation failure in a HVDC system according to any one of claims 1 to 7, characterized in that: The decision binary tree adopts the CART decision classifier.

9. The method for diagnosing commutation failure in a HVDC system according to any one of claims 1 to 7, characterized in that: The kernel function used in the support vector machine is a Gaussian radial basis kernel function.

10. A high voltage direct current transmission system commutation failure fault diagnosis system, comprising a processor, characterized in that: The processor is configured to execute a computer program to implement the steps of the method for diagnosing commutation failure in a high-voltage direct current transmission system according to any one of claims 1 to 9.

Images