Power Electronics AC Tether Protection Method Based on Current Trajectory Recognition

Abstract

A power electronic AC tie-line protection method based on current trajectory recognition includes: collecting three-phase currents at both ends of the AC tie-line, calculating the original differential current and virtual differential current of each phase, and forming a dual differential current image for each phase; acquiring dual differential current images under different grid operating conditions to form a dataset, and preprocessing the dataset; feeding the dataset into an improved convolutional neural network model for training and testing; forming a dual differential current image of the three-phase currents at both ends of an unknown AC tie-line, feeding it into the trained improved convolutional neural network model, and determining whether an intra-zone fault has occurred on the AC tie-line. This method utilizes deep learning to extract discriminative information from complex fault data affected by multiple variables, which is beneficial for accurately determining the fault type. It has significant implications and broad application prospects for improving the intelligence level of relay protection systems and contributing to the safe and stable operation of power systems.

Description

Technical fields:

[0001] This invention relates to the field of power system relay protection technology, and in particular to a power electronic power supply AC tie line protection method based on current trajectory identification. Background technology:

[0002] The integration of high-proportion renewable energy and energy storage power stations into the grid via power electronic devices, coupled with flexible DC transmission technology promoting renewable energy consumption and enabling long-distance, high-capacity power transmission, has profoundly changed the operating characteristics of the power system. Due to limitations imposed by the overcurrent capacity of semiconductor devices and the high controllability of power electronic converters, power electronic power supplies exhibit unique fault characteristics. Traditional relay protection schemes designed for the fault characteristics of traditional synchronous generators are no longer applicable, seriously threatening the stability of the power system and posing a significant challenge to traditional relay protection schemes. Therefore, it is urgent to propose a protection scheme specifically for the characteristics of circuit faults under these new conditions to ensure the large-scale integration of renewable energy and improve the operational reliability of the new power system. Summary of the Invention:

[0003] The purpose of this invention is to design a power electronic AC tie-line protection method based on current trajectory identification to ensure the large-scale integration of new energy sources and improve the operational reliability of new power systems.

[0004] A power electronic power supply AC tie-line protection method based on current trajectory identification includes the following steps:

[0005] Step 1: Collect the three-phase current at both ends of the AC tie line using a current transformer, wherein the two ends of the AC tie line are the power electronic power supply side and the grid side, respectively.

[0006] Step 2: Calculate the initial differential current i for each phase. dx Where x = a, b, c represent phases A, B, and C, respectively; the original differential current of each phase is delayed by one-quarter of the power frequency cycle to obtain the virtual differential current i of each phase. ddx Where T0 is the power frequency period;

[0007] Step 3: Convert the original differential current i of each phase dx and the virtual differential current i of each phase ddx The dynamic trajectory of the dual differential current for each phase is constructed by projecting it onto the x-axis and y-axis respectively. The projection method is as follows: at each time step, i... dx The value is the x-coordinate, i ddxThe value is the vertical axis. When the sampling frequency is 4kHz, N=80, that is, 80 points are sampled in one power frequency cycle, that is, one current data is collected every 0.25ms. The three-phase current data at both ends of the AC tie line collected every 0.25ms are projected onto the x-axis and y-axis as the horizontal and vertical axes.

[0008] Step 4: Starting from T0 / 4 after any fault occurs, record the dynamic trajectory of the dual differential current of each phase for one power frequency cycle to form a dual differential current image of each phase, where T0 is the power frequency cycle.

[0009] Step 5: Acquire dual differential current images under different power grid operating conditions to form a dataset, and label the dual differential current images in the dataset. The label for dual differential current images with intra-zone faults is 1, and the label for dual differential current images with non-intra-zone faults is 0. The different operating conditions include different fault conditions, different operating modes of flexible DC converter stations, different states of energy storage power stations, current transformer saturation, and abnormal data. The data in the dataset used for training is simulation data, and fault occurrence points can be artificially set.

[0010] Step 6: Preprocess the dataset: unify the coordinate axis range, unify the image size of the double differential current images, convert the unified double differential current images into grayscale images, normalize the grayscale images, and perform data augmentation by translation and adding Gaussian noise to generate more equivalent data.

[0011] Step 7: Improve the convolutional neural network model based on channel attention mechanism and decaying learning rate. Divide the preprocessed dataset into training set and test set in a ratio of 8:2. Feed the training set into the improved convolutional neural network model. Through forward propagation and backward propagation, continuously update the network parameters and minimize the output error to obtain the trained improved convolutional neural network model. Then, use the test set to test the trained improved convolutional neural network model.

[0012] Step 8: Collect the three-phase currents at both ends of the unknown AC tie line and perform real-time fault monitoring on the unknown AC tie line. Generate a double differential current image for each phase of the unknown AC tie line. After preprocessing the double differential current images, feed them into the trained improved convolutional neural network model to determine whether an in-zone fault has occurred in the AC tie line and identify the faulty phase. Faults occurring between the relay protection devices at both ends of the AC tie line are called in-zone faults, and vice versa. The double differential current image of an in-zone fault is significantly different from the images when the line is normal and when an out-of-zone fault occurs. If the improved convolutional neural network model outputs 1, it means that an in-zone fault has occurred on this line, and the line needs to be disconnected for protection. If the improved convolutional neural network model outputs 0, it means that a non-in-zone fault has occurred, so it can be considered that no fault has occurred on this line, i.e., no in-zone fault has occurred. Since a double differential current image has been generated for each phase of the three-phase current, it is possible to determine which phase has a fault.

[0013] Preferably, the power electronic power source includes a flexible DC converter station, an energy storage power station, and a new energy power station.

[0014] Preferably, the new energy power station includes photovoltaic power stations and wind farms.

[0015] Preferably, the original differential current of each phase is the sum of the currents of each phase on the power electronic power supply side and the grid side, and the specific calculation method is as follows:

[0016] i dx (k)=i mx (k)+i nx (k)

[0017] i ddx (k)=i dx (kN / 4);

[0018] Where x = a, b, c represent phases A, B, and C respectively, and N is the number of sampling points in one power frequency cycle. For a power system with a power frequency of 50 Hz, when the sampling frequency is 4 kHz, N = 80, i mx (k) and i nx (k) represents the sampling current at the kth sampling point on the power electronic power supply side and the grid side of the AC line, respectively.

[0019] Preferably, for a power system with a power frequency of 50Hz, the power frequency period T0 = 20ms.

[0020] Preferably, the fault conditions include different fault types, fault distances, and transition resistances; the flexible DC converter station operation modes include rectification mode and inverter mode; and the energy storage station status includes charging and discharging.

[0021] The new generation of artificial intelligence technology, represented by deep learning, has strong advantages in data processing, with strong nonlinear fitting and feature expression capabilities. The AC tie-line protection method for power electronic power supplies based on current trajectory recognition designed in this invention combines deep learning with relay protection. It uses deep learning to extract discrimination information from complex fault data that is affected by multiple variables, which is conducive to accurately judging the fault type. This method has great significance and broad application prospects for improving the intelligence level of relay protection system and helping the safe and stable operation of power system. Attached image description:

[0022] Appendix Figure 1 It is a simplified structural diagram of a power electronic power supply system.

[0023] Appendix Figure 2 This is a schematic diagram illustrating the principle of dynamic trajectory formation for dual differential currents.

[0024] Appendix Figure 3 This is a diagram of the structure of an improved convolutional neural network model.

[0025] Appendix Figure 4 This is a flowchart of a power electronic AC tie-line protection method based on current trajectory identification.

[0026] Figure 1 In the middle: Power electronic power supply 1, Flexible DC converter station 11, Energy storage power station 12, Photovoltaic power station 13, Wind farm 14, DC side 111, Photovoltaic 131, Back-to-back converter 141, Transmission line 2, Synchronous grid 3, E N Z N R1 and R2 are the equivalent electromotive force and equivalent internal impedance of the synchronous grid, respectively, and the relay protection devices on the power electronic power supply side (M) and grid side (N) of the AC tie line, respectively. Figure 2 In the equation, the coordinates of the point at time t1 are (i dx (t1), i ddx (t1)); Figure 3 In the diagram: Channel attention module 5, Conv represents the convolutional layer, f is the number of convolutional kernels; Maxpool represents the pooling layer, s represents the stride, FC represents the fully connected layer, Channel attention module represents the channel attention module, H, W, and C represent the length, width, and number of channels of the dual differential current image, and GAP represents global average pooling. Detailed implementation method:

[0027] The following is in conjunction with the instruction manual. Figure 1 ~Attached Figure 4 A preferred embodiment of the AC tie-line protection method for power electronic power supplies based on current trajectory identification provided by the present invention will be described.

[0028] A power electronic power supply AC tie-line protection method based on current trajectory identification includes the following steps:

[0029] Step 100: Collect the three-phase current at both ends of the AC interconnection line through a current transformer. The two ends of the AC interconnection line are the power electronic power supply side and the grid side, respectively. The power electronic power supply includes flexible DC converter stations, energy storage power stations and new energy power stations. The new energy power stations include photovoltaic power stations and wind farms.

[0030] Step 110: Calculate the original differential current i for each phase. dx The original differential current of each phase is delayed by one-quarter of the power frequency cycle to obtain the virtual differential current i of each phase. ddx Where T0 is the power frequency period, the specific calculation method is as follows:

[0031] i dx (k)=i mx (k)+i nx (k)

[0032] i ddx (k)=i dx (kN / 4);

[0033] Where x = a, b, c represent phases A, B, and C respectively, and N is the number of sampling points in one power frequency cycle. For a power system with a power frequency of 50 Hz, when the sampling frequency is 4 kHz, N = 80, i mx (k) and i nx (k) represents the sampling current at the kth sampling point on the power electronic power supply side and the grid side of the AC line, respectively;

[0034] Step 120: Convert the original differential current i of each phase... dx and the virtual differential current i of each phase ddx The dynamic trajectory of the dual differential current for each phase is constructed by projecting it onto the x-axis and y-axis respectively. The projection method is as follows: i is taken every 0.25ms. dx The value is the x-coordinate, i ddx The value is the ordinate, (i dx i ddx The trajectory formed by the moving point is the dynamic trajectory of the double differential current;

[0035] Step 130: Starting from T0 / 4 after any fault occurs, the dynamic trajectory of the dual differential current of each phase is continuously recorded for one power frequency cycle to form a dual differential current image of each phase. In a power system with a power frequency of 50Hz, the power frequency cycle T0 = 20ms.

[0036] Step 140: Acquire dual differential current images under different grid operating conditions to form a dataset, and label the dual differential current images in the dataset. The label for dual differential current images with in-zone faults is 1, and the label for dual differential current images with non-in-zone faults is 0. The different operating conditions include different fault conditions, different operating modes of flexible DC converter stations, different states of energy storage stations, current transformer saturation, and abnormal data. The fault conditions include different fault types, fault distances, and transition resistances. The operating modes of flexible DC converter stations include rectification mode and inverter mode. The states of energy storage stations include charging and discharging.

[0037] Step 150: Preprocess the dataset: unify the coordinate axis range, unify the image size of the double differential current image, convert the unified double differential current image into a grayscale image, normalize the grayscale image, and perform data augmentation by translation and adding Gaussian noise to generate more equivalent data.

[0038] Step 160: Improve the convolutional neural network model based on channel attention mechanism and decaying learning rate. Divide the preprocessed dataset into training set and test set in a ratio of 8:2. Feed the training set into the improved convolutional neural network model. Continuously update the network parameters through forward propagation and backward propagation to minimize the output error and obtain the trained improved convolutional neural network model. Test the trained improved convolutional neural network model using the test set.

[0039] This paper incorporates a channel attention mechanism into traditional convolutional neural networks. GAP stands for Global Average Pooling, which averages the values ​​of each channel in the feature map. This preserves feature map information while reducing its size from H×W×C to 1×1×C. The output of the previous layer is connected to the global average pooling layer, fully connected layer, and sigmoid activation function to generate a 1×1×C weight matrix. This weight matrix is ​​then multiplied by the output of the previous layer to become the input feature map for the next layer.

[0040] The expression for the sigmoid activation function is:

[0041]

[0042] x and f(x) represent the input and output, respectively;

[0043] Decaying learning involves selecting a relatively large learning rate in the early stages of network training to accelerate convergence. As the number of iterations increases, the learning rate is gradually decreased to prevent oscillations when the network converges to the vicinity of the optimum, thus improving convergence to the optimal solution. The formula for the learning rate α is as follows:

[0044]

[0045] Where α0 represents the initial learning rate, epoch represents the number of iterations, and the learning rate begins to decay when the number of iterations exceeds the set value N;

[0046] Step 170: Collect the three-phase currents at both ends of the unknown AC tie line and perform real-time fault monitoring on the unknown AC tie line. Generate a double differential current image for each phase of the unknown AC tie line. After preprocessing the double differential current images, feed them into the trained improved convolutional neural network model to determine whether an in-zone fault has occurred in the AC tie line and identify the faulty phase. Faults occurring between the relay protection devices at both ends of the AC tie line are called in-zone faults, and vice versa. The double differential current image of an in-zone fault is significantly different from the images when the line is normal and when an out-of-zone fault occurs. If the improved convolutional neural network model outputs 1, it means that an in-zone fault has occurred on this line, and the line needs to be disconnected for protection. If the improved convolutional neural network model outputs 0, it means that a non-in-zone fault has occurred. Therefore, it can be considered that no fault has occurred on this line, i.e., no in-zone fault has occurred. Since a double differential current image has been generated for each phase of the three-phase current, it is possible to determine which phase has a fault.

[0047] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements, substitutions, modifications and refinements can be made without departing from the principles and spirit of the present invention, and these improvements, substitutions, modifications and refinements should also be considered within the scope of protection of the present invention.

Claims

1. A power electronic power supply AC tie-line protection method based on current trajectory recognition, characterized in that, The method comprises the following steps: Step one: collecting three-phase currents at both ends of the AC tie line through a current transformer, wherein the AC tie line is connected to a power electronic power supply side and a power grid side respectively; Step two: calculate the original differential current i of each phase respectively dx Wherein, x=a, b, c respectively represent A phase, B phase and C phase; delay the original differential current of each phase by one fourth of the power frequency cycle to obtain the virtual differential current i of each phase ddx ​ Step three: project the original differential current i dx and the virtual differential current i ddx of each phase respectively to x-axis and y-axis to construct the dynamic trajectory of double differential current of each phase, wherein the projection method is: taking the value of original differential current i dx as the horizontal coordinate and the value of virtual differential current i ddx as the vertical coordinate at each time. Step four: starting from T0 / 4 after any fault occurs, the dynamic trajectory of the double differential current of each phase for one power frequency cycle forms a double differential current image of each phase, wherein T0 is a power frequency cycle; Step five: obtaining double differential current images under different operating conditions of the power grid to form a data set, wherein the different operating conditions include different fault conditions, different operating modes of the flexible HVDC converter station, different states of the energy storage power station, current transformer saturation, and abnormal data; Step six: preprocessing the data set: unifying the coordinate axis range, unifying the picture size of the double differential current image, converting the double differential current image of the unified size into a grayscale image, and performing data enhancement after normalizing the grayscale image; Step seven: based on the improved convolutional neural network model with channel attention mechanism and decay learning rate, the preprocessed data set is divided into a training set and a test set in a ratio of 8:2, the training set is fed into the improved convolutional neural network model, the network parameters are updated through forward propagation and back propagation, the output error is minimized, and a trained improved convolutional neural network model is obtained, and the test set is used to test the trained improved convolutional neural network model; Step eight: collecting three-phase currents at both ends of the unknown AC tie line and performing real-time fault monitoring on the unknown AC tie line to generate a double differential current image of each phase of the unknown AC tie line, preprocessing the double differential current image, and feeding it into the trained improved convolutional neural network model to determine whether the unknown AC tie line has an internal fault and the fault phase.

2. The current trajectory recognition based power electronic power supply ac tie-line protection method according to claim 1, wherein, The power electronic power supply side includes a flexible HVDC converter station, an energy storage power station, and a new energy station.

3. The current trajectory recognition based power electronic power supply AC tie-line protection method according to claim 2, wherein, The new energy station includes a photovoltaic power station and a wind power station.

4. The current trajectory recognition based power electronic power supply ac tie-line protection method according to claim 1, wherein, The original differential current i of each phase dx is the sum of the currents of each phase on the power electronics supply side and on the grid side.

5. The current trajectory recognition based power electronic power supply ac tie-line protection method according to claim 1, wherein, The power frequency of the power system is 50Hz, and the power frequency cycle T0 is 20ms.

6. The current trajectory recognition based power electronic supply AC tie-line protection method according to claim 1, wherein, The different fault conditions include different fault types, fault distances, and transition resistances; the different operating modes of the flexible HVDC converter station include rectification mode and inversion mode; and the different states of the energy storage power station include charging and discharging.

Images