Wind power transmission line fault detection method

Abstract

The invention provides a wind power transmission line fault detection method, which comprises the steps of building a doubly-fed fan model, setting a fault on an output line and analyzing fault characteristics, and collecting fault transient current data and obtaining a state transition matrix according to different transition resistances and fault distances; and finally, fault features are extracted, and fault detection is carried out by using the built and trained CNN-BiLSTM model. According to the fault detection scheme provided by the invention, the adopted data is the original transient current data, the extraction is simple, different fault areas can be accurately judged by using the difference of the waveform characteristics of different fault positions, and the fault data is acquired by only one protection device on one power transmission line in the scheme, so that the fault detection efficiency is improved. According to the invention, the method can achieve the recognition of the fault region of the whole section of line, is low in cost and is high in efficiency, the proposed fault detection scheme avoids the complex calculation and threshold setting, and the proposed deep learning method has the capability of real-time monitoring and rapid fault information capturing.

Description

Technical Field

[0001] This application belongs to the field of power technology, and in particular relates to a method for detecting faults in wind power transmission lines. Background Technology

[0002] Currently, environmental pollution and energy transition have become two major issues facing social development. The recoverable reserves of traditional fossil fuels, such as coal and oil, are decreasing day by day, and their use has also seriously polluted the human living environment.

[0003] Wind power is a renewable energy source that can reduce environmental pollution and greenhouse gas emissions. The principle of wind power generation is to use wind power to drive the blades of a windmill to rotate, thereby generating electricity. Wind power generation does not produce pollution or consume fossil fuels, making it an environmentally friendly energy source.

[0004] Currently, wind power, as a relatively mature form of renewable energy with the potential for large-scale development, has a significantly lower minimum cost than traditional fossil fuels, resulting in good economic benefits. With technological advancements, more wind farms have been built. The large-scale transmission of wind power into the power system inevitably impacts the system, potentially leading to wind curtailment. To improve the availability, reliability, and economy of the power system, in addition to further increasing wind turbine capacity, in-depth research is urgently needed on fault diagnosis technology for large-scale wind power transmission lines to improve transmission efficiency and system stability. Summary of the Invention

[0005] This application provides a method for detecting faults in wind power transmission lines. It can not only collect information on the transmission line in real time, but also avoid the problems of difficulty in extracting fault features, complex calculations for protection actions, and threshold setting. It solves the problems of complicated manual threshold setting and difficulty in extracting fault features in traditional protection.

[0006] This application provides a method for detecting faults in wind power transmission lines, including:

[0007] Build a doubly fed wind turbine model in PSCAD / EMTDC;

[0008] Different types of faults are set at different locations on the power transmission line of the doubly fed wind turbine model, and the fault transient current simulation data are collected through the protection device.

[0009] A Markov model is constructed, and the simulation data of the fault transient current is input into the Markov model to obtain the state transition matrix;

[0010] A CNN-BiLSTM was constructed, and the fault feature information of the state transition matrix was extracted using the CNN-BiLSTM.

[0011] The fault feature information is used to train the CNN-BiLSTN network and set its parameters;

[0012] The trained CNN-BiLSTM is used to detect different faults, and the fault detection results are obtained.

[0013] A fault detection method for wind power transmission lines is proposed. This method leverages the differences in amplitude and abrupt changes of the transient current waveform at different fault locations, combined with deep learning networks, to create a large-scale fault detection method for wind power transmission lines based on Markov and CNN-BiLSTM. First, transient current data is collected from protection devices on the transmission line and used as input samples for a Markov model to obtain the state transition matrix. Second, a constructed CNN-BiLSTM network is used to extract fault features from the state transition matrix and train the network. The network's structural parameters are determined through parameter comparison experiments. Finally, the offline-trained CNN-BiLSTM network is used to detect faults on the transmission line, triggering corresponding protection actions.

[0014] In one feasible implementation, the specific steps of building the doubly fed wind turbine model in PSCAD / EMTDC are as follows:

[0015] In the single-unit model of a doubly fed wind turbine, the motor stator is connected to the power grid via a grid-connected switch, and the motor rotor is connected to the power grid via a machine-side converter and a grid-side converter.

[0016] In one feasible implementation, the method of setting different types of faults at different locations on the power transmission line of the doubly-fed wind turbine model includes:

[0017] A-phase short circuit and ABC three-phase short circuit were set at four different locations on the output line. Simulations were performed based on different transition resistances and fault distances. The transition resistance step size was 5Ω and the fault distance step size was 50km. There were a total of 8 faults, and 100 sets of data were simulated for each fault, resulting in a total of 800 sets of data.

[0018] In one feasible implementation, the method for extracting fault feature information from the state transition matrix using CNN-BiLSTM includes:

[0019] CNN-BiLSTM performs three convolutional operations on the input feature sequence to extract features;

[0020] The output of the convolutional layer is used as the input of the BiLSTM to perform two layers of BiLSTM operations in order to capture the temporal information in the sequence data.

[0021] The output of BiLSTM is processed through a fully connected layer to obtain the final output result.

[0022] In one feasible implementation, the method for training the CNN-BiLSTN network using the fault feature information is as follows:

[0023] 80% of the fault data was extracted as training samples and trained using CNN-BiLSTM to learn and memorize the fault features of 8 different fault locations. The remaining 20% ​​of the fault data was used as validation samples.

[0024] For CNN modules, the appropriate kernel size is key to extracting transient current features of faults, which is beneficial for capturing local details of transient current signals. For BiLSTM networks, the number of neurons greatly affects the model's recognition accuracy and efficiency.

[0025] In one feasible implementation, CNN-BiLSTM is used to learn and memorize the fault features of the fault data. The learning and memorization process is as follows:

[0026] The role of the CNN layer is to extract spatial coupling features from the acquired fault transient current data in blocks, thereby significantly enhancing the deep spatial features. Its calculation formula is as follows:

[0027]

[0028] In the formula, g(·) represents the ReLU activation function; x represents the current network layer number; This indicates that a convolution operation is being performed; K represents the output of a certain layer; W represents the weights of neurons in a certain layer; b is the bias.

[0029] BiLSTM consists of four layers: an input layer, a forward LSTM layer, a backward LSTM layer, and an output layer. The number of neurons in the input layer is determined by the amount of input, and the number of neurons in the output layer is determined by the amount of output.

[0030] The role of the fully connected layer is to abstract and combine the features extracted by the convolutional layer, mapping the distributed features extracted by the convolutional layer to the sample label space, that is, outputting a fixed-length feature vector, the calculation formula of which is:

[0031]

[0032] In the formula, K represents the output of a certain layer; x represents the current network layer number; g(·) represents the ReLU activation function; W represents the weight of a neuron in a certain layer; and b is the bias.

[0033] In one feasible implementation, the parameters of the CNN-BiLSTM are set as follows:

[0034] The convolution kernel size is 3×3, and the number of LSTM neurons is set to 100.

[0035] In one feasible implementation, the standardized sample input is X, and the output is denoted as Y = (y1, y2, y3, y4, y5, y6, y7, y8), where y1 to y8 represent 8 different faults.

[0036] This application provides a method for detecting faults in wind power transmission lines. Based on the differences in amplitude information and abrupt changes in the transient current waveforms at different fault locations, and combined with a deep learning network, a large-scale wind power transmission line fault detection method based on Markov and CNN-BiLSTM is proposed. First, transient current data is collected using protection devices on the transmission line, serving as input samples for a Markov model to obtain the state transition matrix. Second, a constructed CNN-BiLSTM network is used to extract fault features from the state transition matrix and train the network. The network's structural parameters are determined through parameter comparison experiments. Finally, the offline-trained CNN-BiLSTM network is used to detect faults on the transmission line, triggering corresponding protection actions.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are:

[0038] (1) The fault detection scheme provided in this application uses raw transient current data, which is easy to extract. By utilizing the differences in waveform characteristics at different fault locations, different fault areas can be accurately identified.

[0039] (2) This scheme can identify the fault area of ​​the entire line by relying on only one protection device to obtain fault data on a single transmission line, which has the advantages of low cost and high efficiency.

[0040] (3) The proposed fault detection scheme avoids complex calculations and threshold tuning.

[0041] (4) The proposed deep learning method has the ability to monitor and quickly capture fault information in real time. Attached Figure Description

[0042] Figure 1 This is a flowchart of the wind power transmission line fault detection method provided in this application;

[0043] Figure 2 This is a protection flowchart for wind power transmission line fault detection methods;

[0044] Figure 3 This is a schematic diagram of the topology of a large-scale wind power transmission line;

[0045] Figure 4This is a schematic diagram of the CNN-BiLSTM structure. Detailed Implementation

[0046] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of this application.

[0047] With technological advancements, more wind farms have been built. The large-scale transmission of wind power into the system inevitably has some impact on the system, and may even lead to "wind curtailment." In order to improve the availability, reliability, and economy of the power system, in addition to further increasing the capacity of wind turbines, it is also necessary to conduct in-depth research on fault diagnosis technology for large-scale wind power transmission lines in order to improve the transmission efficiency of wind power generation and the stability of the system.

[0048] To address the challenges of complex manual threshold settings and difficult fault feature extraction in traditional protection systems, this paper proposes a large-scale wind power transmission line fault detection method based on Markov and CNN-BiLSTM, taking into account the differences in amplitude and abrupt changes in fault transient current waveforms at different fault locations. First, fault transient current data is collected using protection devices on the transmission line and used as input samples for a Markov model to obtain the state transition matrix. Second, a constructed CNN-BiLSTM network is used to extract fault features from the state transition matrix and train the network. The network's structural parameters are determined through parameter comparison experiments. Finally, the offline-trained CNN-BiLSTM network is used to detect faults on the transmission line, triggering corresponding protection actions.

[0049] The following detailed description, in conjunction with the accompanying drawings, illustrates the specific structure of the wind power transmission line fault detection method provided in this application.

[0050] Reference Figures 1-4 As shown in the figure, this application provides a method for detecting faults in wind power transmission lines, including the following steps:

[0051] S100: First, build a doubly fed wind turbine model in PSCAD / EMTDC.

[0052] PSCAD / EMTDC is a software tool for power system simulation, developed by Manitoba HVDC Research Centre. It provides a comprehensive simulation environment for analyzing and designing various components and systems within power systems.

[0053] PSCAD / EMTDC can simulate various phenomena in power systems, such as power flow, fault analysis, stability analysis, and control system design. It can also simulate various types of power electronic equipment, such as inverters, rectifiers, filters, and transformers.

[0054] PSCAD / EMTDC boasts powerful modeling and simulation capabilities, and is easy to use and customize. It can integrate with other power system software tools such as MATLAB, Simulink, and ETAP.

[0055] In summary, PSCAD / EMTDC is a very useful power system simulation tool, widely used in power system design, analysis, and optimization.

[0056] The specific steps for building a doubly fed wind turbine model in PSCAD / EMTDC are as follows:

[0057] In the single-unit model of a doubly fed wind turbine, its stator is connected to the power grid via a grid-connected switch, and the motor rotor is connected to the power grid via a machine-side converter and a grid-side converter.

[0058] S200: Different types of faults are set at different locations on the doubly fed wind turbine output line, and the fault transient current simulation data are collected using the protection device.

[0059] Methods for setting different types of faults at different locations on the power transmission line of the doubly fed wind turbine model include:

[0060] like Figure 3 As shown, A-phase short circuit and ABC three-phase short circuit were set at four different locations (f1, f2, f3, f4) on the output line, respectively. Simulation was performed according to the different transition resistance and fault distance. The transition resistance step size was 5Ω and the fault distance step size was 50km. There were a total of 8 faults, and 100 sets of data were simulated for each fault, resulting in a total of 800 sets of data.

[0061] S300: Build a Markov model and input the fault transient current simulation data into the Markov model to obtain the state transition matrix.

[0062] The CNN-BiLSTM Markov model is a deep learning model that combines a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM) for processing sequential data.

[0063] Convolutional Neural Networks (CNNs) are a commonly used image recognition model that extracts features and classifies images using convolutional kernels. In sequence data processing, CNNs can be used for feature extraction and classification of sequential data.

[0064] Bidirectional Long Short-Term Memory (BiLSTM) is a special type of recurrent neural network (RNN) that possesses bidirectional memory capabilities, enabling it to better handle long-term dependencies in sequential data. BiLSTM can be used for modeling and predicting sequential data.

[0065] A Markov model is a statistical model used to describe the state transition patterns of a stochastic process. In sequence data processing, Markov models can be used to model and predict sequence data.

[0066] The CNN-BiLSTM Markov model combines CNN and BiLSTM. It uses CNN to extract features from sequence data, then uses BiLSTM to model and predict the sequence data, and finally uses a Markov model to optimize the prediction results. This model can be used to process various types of sequence data, such as natural language processing, speech recognition, and time series prediction.

[0067] Before obtaining the state transition matrix, a Markov model must first be constructed based on the actual problem to be solved. The state transition matrix describes the probability of the system transitioning from one state to another. This invention has two states: transitioning from a normal state to a single-phase fault and a three-phase fault, respectively. In this case, the state transition matrix is ​​a second-order matrix. This invention mainly utilizes the Markov model's ability to monitor fault signals in real time and quickly acquire fault information, enabling a good response to fault states on the line and obtaining high-quality data.

[0068] S400: Use the already built CNN-BiLSTM to extract the feature information of the above state transition matrix.

[0069] like Figure 4 As shown, the specific features extracted from the state transition matrix include:

[0070] S410: CNN-BiLSTM first performs three convolutional operations on the input feature sequence to extract features;

[0071] S420: The output of the convolutional layer is used as the input of BiLSTM to perform two layers of BiLSTM operation to capture temporal information in sequence data. In this way, CNN-BiLSTM greatly reduces the computational cost and the risk of overfitting while extracting features.

[0072] S430: Finally, the BiLSTM output is processed through a fully connected layer to obtain the final output result. CNN-BiLSTM has shown excellent performance in sequence data processing tasks.

[0073] The corresponding fault features are extracted using the methods described above. The network structure of CNN-BiLSTM is shown in the figure below. Figure 3 As shown.

[0074] S500: The CNN-BiLSTN network was trained and its parameters were set using simulation data. 80% of the fault data was used as training samples. The CNN-BiLSTM network was trained to learn and memorize the fault features of eight different fault locations. The remaining 20% ​​of the fault data was used as validation samples. For the CNN module, an appropriate convolution kernel size is key to extracting transient current features from faults, which is beneficial for capturing local details of transient current signals. For the BiLSTM network, the number of neurons greatly affects the model's recognition accuracy and efficiency.

[0075] S600: Standardize the simulation data and use the previously trained CNN-BiLSTM to detect different faults to obtain fault detection results.

[0076] Due to significant differences among the training data, the fault features are unevenly distributed. Therefore, to maximize the diversity of fault features and improve the training accuracy of the network, the training data needs to be standardized. The previously trained CNN-BiLSTM is then used to detect different faults, yielding fault detection results.

[0077] This application provides a fault detection method for wind power transmission lines. It proposes a large-scale fault detection method for wind power transmission lines based on Markov and CNN-BiLSTM, taking into account the differences in amplitude and abrupt change times of fault transient current waveforms at different fault locations, and combining this with a deep learning network. First, fault transient current data is collected using protection devices on the transmission line, serving as input samples for a Markov model to obtain the state transition matrix. Second, a constructed CNN-BiLSTM network is used to extract fault features from the state transition matrix and train the network. The network's structural parameters are determined through parameter comparison experiments. Finally, the offline-trained CNN-BiLSTM network is used to detect faults on the transmission line, triggering corresponding protection actions.

[0078] The main process of this wind power transmission line fault detection method is as follows:

[0079] A doubly fed wind turbine model was built, and faults were set on the output line and the fault characteristics were analyzed. Then, based on the different transition resistance and fault distance, the transient current data of the fault was collected by the protection device and the state transition matrix was obtained. Finally, the fault features were extracted and the fault detection was performed using the CNN-BiLSTM model that had been built and trained.

[0080] In some embodiments, CNN-BiLSTM is used to learn and memorize the fault features of fault data. The learning and memorization process is as follows:

[0081] The role of the CNN layer is to extract spatial coupling features from the acquired fault transient current data by segmenting it into blocks, thus significantly enhancing the deep spatial features. Its calculation formula is:

[0082]

[0083] In the formula, g(·) represents the ReLU activation function; x represents the current network layer number; This indicates that a convolution operation is being performed; K represents the output of a certain layer; W represents the weights of neurons in a certain layer; b is the bias.

[0084] BiLSTM consists of four layers: an input layer, a forward LSTM layer, a backward LSTM layer, and an output layer. The number of neurons in the input layer is determined by the input quantity, and the number of neurons in the output layer is determined by the output quantity.

[0085] The role of a fully connected layer is to abstract and combine the features extracted by the convolutional layers, mapping the distributed features extracted by the convolutional layers to the sample label space, i.e., outputting a fixed-length feature vector. Its calculation formula is:

[0086]

[0087] In the formula, K represents the output of a certain layer; x represents the current network layer number; g(·) represents the ReLU activation function; W represents the weight of a neuron in a certain layer; and b is the bias.

[0088] To achieve fault detection, this paper designs eight output results.

[0089] In some embodiments, the final parameters of CNN-BiLSTM are set as follows: the kernel size is 3×3 and the number of LSTM neurons is set to 100.

[0090] In some embodiments, the standardized sample input is X, and the output is denoted as Y = (y1, y2, y3, y4, y5, y6, y7, y8), where y1 to y8 represent 8 different faults.

[0091] It should be noted that the numerical values ​​and ranges involved in the embodiments of this application are approximate values. Due to the influence of the manufacturing process, there may be a certain range of errors, which can be considered negligible by those skilled in the art.

[0092] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0093] It is readily understood that, based on the several embodiments provided in this application, those skilled in the art can combine, split, or reorganize the embodiments of this application to obtain other embodiments, none of which exceed the protection scope of this application.

[0094] The above detailed embodiments further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A method for detecting faults in wind power transmission lines, characterized in that: include; Build a doubly fed wind turbine model in PSCAD / EMTDC; Different types of faults are set at different locations on the power transmission line of the doubly fed wind turbine model, and the fault transient current simulation data are collected through the protection device. A Markov model is constructed, and the simulation data of the fault transient current is input into the Markov model to obtain the state transition matrix; A CNN-BiLSTM was constructed, and the fault feature information of the state transition matrix was extracted using the CNN-BiLSTM. The fault feature information is used to train the CNN-BiLSTN network and set its parameters; The trained CNN-BiLSTM is used to detect different faults, and the fault detection results are obtained.

2. The method for detecting faults in wind power transmission lines according to claim 1, characterized in that: The specific steps for building a doubly fed wind turbine model in PSCAD / EMTDC are as follows: In the single-unit model of a doubly fed wind turbine, the motor stator is connected to the power grid via a grid-connected switch, and the motor rotor is connected to the power grid via a machine-side converter and a grid-side converter.

3. The wind power transmission line fault detection method according to claim 1, characterized in that: The method of setting different types of faults at different locations on the power transmission line of the doubly fed wind turbine model includes: A-phase short circuit and ABC three-phase short circuit were set at four different locations on the output line. Simulations were performed based on different transition resistances and fault distances. The transition resistance step size was 5Ω and the fault distance step size was 50km. There were a total of 8 faults, and 100 sets of data were simulated for each fault, resulting in a total of 800 sets of data.

4. The method for detecting faults in wind power transmission lines according to claim 1, characterized in that: Methods for extracting fault feature information from the state transition matrix using CNN-BiLSTM include: CNN-BiLSTM performs three convolutional operations on the input feature sequence to extract features; The output of the convolutional layer is used as the input of the BiLSTM to perform two layers of BiLSTM operations in order to capture the temporal information in the sequence data. The output of BiLSTM is processed through a fully connected layer to obtain the final output result.

5. The wind power transmission line fault detection method according to claim 3, characterized in that: The method for training the CNN-BiLSTN network using the aforementioned fault feature information is as follows: 80% of the fault data was extracted as training samples and trained using CNN-BiLSTM to learn and memorize the fault features of 8 different fault locations. The remaining 20% ​​of the fault data was used as validation samples. For CNN modules, the appropriate kernel size is key to extracting transient current features of faults, which is beneficial for capturing local details of transient current signals. For BiLSTM networks, the number of neurons greatly affects the model's recognition accuracy and efficiency.

6. The wind power transmission line fault detection method according to claim 5, characterized in that: The CNN-BiLSTM method learns and memorizes the fault features of fault data. The learning and memorization process is as follows: The role of the CNN layer is to extract spatial coupling features from the acquired fault transient current data in blocks, thereby significantly enhancing the deep spatial features. Its calculation formula is as follows: In the formula, g(·) represents the ReLU activation function; x represents the current network layer number; This indicates that a convolution operation is being performed; K represents the output of a certain layer; W represents the weights of neurons in a certain layer; b is the bias. BiLSTM consists of four layers: an input layer, a forward LSTM layer, a backward LSTM layer, and an output layer. The number of neurons in the input layer is determined by the amount of input, and the number of neurons in the output layer is determined by the amount of output. The role of the fully connected layer is to abstract and combine the features extracted by the convolutional layer, mapping the distributed features extracted by the convolutional layer to the sample label space, that is, outputting a fixed-length feature vector, the calculation formula of which is: K x =g(W x K x-1 +b); In the formula, K represents the output of a certain layer; x represents the current network layer number; g(·) represents the ReLU activation function; W represents the weight of a neuron in a certain layer; and b is the bias.

7. The wind power transmission line fault detection method according to claim 3, characterized in that: The parameters of the CNN-BiLSTM are set as follows: The convolution kernel size is 3×3, and the number of LSTM neurons is set to 100.

Images