Fault detection method and system for double-bus system

Abstract

The invention discloses a fault detection method and system for a double-bus system, and the method comprises the steps: obtaining the current of each branch and the current of a bus coupler on two buses under the double-bus system; when it is judged that a fault occurs in the area, large-difference differential current and braking current under the double-bus system are solved, when it is judged that the fault occurs in the area, small-difference differential current of two buses under the double-bus system is solved, and the large-difference differential current and the small-difference differential current of the two buses are processed; time domain feature data of the large-difference differential current and time domain feature data of the small-difference differential current of the two buses are fused, a BP neural network is trained through a fused data set, and a trained double-bus fault detection model is obtained; and the trained double-bus fault detection model is utilized to analyze and diagnose the time domain characteristics of the double buses, so that the fault occurrence position and the fault type can be judged. According to the method, the specific fault type can be judged by using the algorithm model, and the accuracy and reliability of bus fault detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network, and particularly relates to a fault detection method and system for a double-bus structure. BACKGROUND

[0002] In a power system, as one of the most important components, a bus bar undertakes the important task of collecting and distributing power, and its stable operation directly affects the reliability and safety of the entire power system. There are many different reasons for bus bar failure, such as short circuit, insulation aging, equipment damage, animal contact, and misoperation. When a bus bar fails, it must be handled in time to protect the safety of the power system. Although there are different reasons for bus bar failure, the main concern in fault analysis is to quickly find the type of fault.

[0003] At present, the mainstream bus protection method is through current differential protection, which can disconnect the fault line when a fault occurs, and maintain the power grid to continue to operate until the maintenance personnel repair. However, pure relay protection means cannot effectively give the fault reason, and repair personnel still need to distinguish the fault type and fault reason by themselves. With the continuous development of artificial intelligence technologies such as neural networks and deep learning in the field of fault diagnosis, their good data analysis capabilities can better adapt to the complexity of power system faults and can promote the progress of modern power system fault analysis technology.

[0004] Regarding the protection object as a neural network model is a common method for solving problems by using neural networks. By classifying data under different conditions, different characteristics of different conditions are obtained, and after inputting the data amount, the corresponding condition of the data amount is obtained, so as to guide the protection device to make a judgment whether to act, help the bus protection to better distinguish the position of the fault occurrence, judge the type of the fault occurrence, and improve the accuracy and reliability of the bus fault detection. However, due to the complexity of the characteristic parameters in the bus system, how to select appropriate characteristic parameters, improve the fitting degree of the prediction result and the actual result, and how to process the data of the characteristic parameters to avoid overfitting problem of the neural network model caused by complex calculation, become the difficulties of the application of neural networks in actual bus fault diagnosis. SUMMARY

[0005] The present application provides a fault detection method and system for a double-bus system that can judge the specific fault type and improve the accuracy and reliability of bus fault detection.

[0006] The technical scheme of the present application is as follows: a fault detection method for a double-bus system, comprising: obtaining the branch currents on the two buses and the bus-tie current under the double-bus system; Solving the large difference differential current and the braking current under the double bus system, if the ratio of the large difference differential current to the braking current is greater than or equal to the ratio braking coefficient, it is judged that the fault occurs in the region, and the double bus system fault detection step is entered, otherwise it is judged that the fault occurs outside the region, and the protection does not act; When it is judged that the fault occurs in the region, the small difference differential current of the two buses under the double bus system is solved, and the large difference differential current and the small difference differential current of the two buses are processed to obtain the time domain characteristic data; The time domain characteristic data of the large difference differential current and the small difference differential current of the two buses is fused based on the DS evidence theory to obtain the double bus system time domain characteristic training data, and the trained double bus fault detection model is obtained by training the data set after fusion on the BP neural network; The trained double bus fault detection model is used to analyze and diagnose the time domain characteristics of the double bus, so as to judge the position and type of the fault.

[0007] The large difference differential current, the small difference differential current and the braking current solving method comprises: The large difference differential current of the double bus After regarding the double bus and all connected branches (including the bus coupler) as a whole, the vector sum of all branch currents, the small difference differential current of the bus And The vector sum of the branch currents of bus 1 and bus 2, the braking current I res The scalar sum of all branch currents and bus coupler currents.

[0008] The region-in and region-out fault discrimination takes the ratio of the large difference differential current of the double bus to the braking current as the discrimination basis, comprising: ; Among them, k set The braking coefficient is usually set to 0.3-0.7, and 0.3 is taken here. When the condition is met, it is judged that the fault occurs in the region, and the subsequent double bus fault detection process is entered, otherwise it is considered that the fault does not occur or occurs outside the region.

[0009] The training data collection and diagnosis process of the double bus system fault detection model comprises: After the large difference differential current of the double bus system and the small difference differential current of the two buses are solved, the time domain basic characteristic parameters a 1- a 7 and auxiliary time domain characteristic parameters a 8- a 10 are obtained; The time-domain characteristic parameters of the three currents are fused using the DS evidence theory to obtain the training data matrix X of the dual-bus fault detection model; The dual-bus fault detection model is constructed primarily using a BP neural network. After training and adjustment, it determines the fault type and location.

[0010] The time-domain characteristic parameters of the large differential current of the busbar and the small differential currents of the two busesbars are calculated as follows: ; In the formula, N Indicates the number of data points. x(n) Indicates waveform, a 1-7 Represents the eigenvalues ​​of a time-domain scalar. a 8 This represents the maximum value of the differential current waveform amplitude at the busbar. a 9 This represents the minimum amplitude of the differential current waveform at the busbar. a 10 This indicates the peak value.

[0011] The dual-bus fault detection model employs Dempster-Shafer Evidence Theory during training to optimize network input, comprehensively reflecting the temporal characteristics of fault waveforms in a dual-bus system. This effectively shortens training time and reduces uncertainty. Multi-feature fusion is implemented based on a generalized form of Dempster's synthesis rules. ; In the formula, m i For the first i The basic probability assignment function for each source of evidence. n This represents the total number of time-domain feature parameters, which is 30 in this case. A To determine the type of fault, A i The fault type supported by local evidence provided for a single feature, where K is the conflict quantity. The feature matrix used for model training is obtained in this way; the feature matrix X of a single input is [ m 1, m 2, m 3… m 10 ] T ,in m These are the characteristic parameters after fusion.

[0012] The dual-bus fault detection model is trained using a backpropagation network (BP neural network), iteratively performing forward and backward propagation. The BP neural network progressively optimizes the weights and biases, ultimately fitting the relationship between fault features and fault types. For the dataset optimized using DS evidence theory (i.e., the training set), the BP neural network employs gradient search to progressively reduce the error, and the transfer function is chosen to be the continuous derivative sigmoid function. ; In the formula, f(x) For the transfer function, exp( x ) is an exponential function, x This is the input to the neuron.

[0013] Secondly, the present invention provides a fault detection system for a dual-bus system, comprising: The data acquisition module is used to acquire the branch current and bus tie current on each of the two buses in a dual-bus system. The judgment module is used to solve the large differential current and braking current in a double bus system. If the ratio of the large differential current to the braking current is greater than or equal to the ratio braking coefficient, it is determined that the fault occurs within the area and the double bus system fault detection step is entered. Otherwise, it is determined that the fault occurs outside the area and the protection does not operate. The time-domain module is used to solve the small differential current of the two buses in a dual-bus system when the fault is determined to occur within the region. It processes the large differential current and the small differential current of the two buses to obtain time-domain characteristic data. The training module is used to fuse the time-domain feature data of the large differential current and the small differential current of the two buses based on the DS evidence theory to obtain the time-domain feature training data of the dual-bus system. The BP neural network is then trained using the fused dataset to obtain the trained dual-bus fault detection model. The detection module is used to analyze and diagnose the time-domain characteristics of the dual busbars using a trained dual busbar fault detection model, thereby determining the location and type of the fault.

[0014] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: This method is based on the acquired large and small differential currents of the busbars in the area to be tested. First, it determines whether the fault is outside or inside the area. Then, based on the determination result, it performs fault detection based on the waveform data of the small differential current of the busbars. The fault detection based on the small differential current of the busbars uses a trained neural network algorithm model to detect the fault type and fault location, thereby obtaining the corresponding detection results and improving the accuracy and reliability of fault detection.

[0015] The training set for the fault detection model uses a time-domain feature dataset, which includes the time-domain feature parameters of the bus large differential current and the bus small differential current during normal operation and when a fault occurs. Based on the training set, the dual-bus fault detection model is trained to determine the optimal parameters of the dual-bus system fault detection model, so that the trained dual-bus system fault detection model has higher prediction accuracy. The fault detection model for a dual-bus system is based on the DS evidence theory to fuse time-domain feature parameters to obtain the network input for training the bus fault model. This avoids the complex calculations and uncertainties of the training process to a certain extent and improves the accuracy of diagnosis. Attached Figure Description

[0016] Figure 1 The diagram shown is a schematic flowchart of a fault detection method for a dual-bus system in one embodiment of the present invention. Figure 2 The diagram shown is a schematic diagram of a dual-bus power system structure in one embodiment of the present invention; Figure 3 The figure shown is a simulation model of a primary system with a dual-busbar structure in one embodiment of the present invention; Figure 4 The figure shown is a differential current waveform diagram of a three-phase short-circuit fault on a bus in one embodiment of the present invention. Figure 5 The diagram shown illustrates the operation of a differential element during a three-phase short-circuit fault on a busbar in one embodiment of the present invention. Figure 6 The figure shown is a mean square error curve of the DS-BP neural network training process in one embodiment of the present invention. Figure 7 The figure shown is a training measurement trend diagram of the DS-BP neural network training process in one embodiment of the present invention; Figure 8 The figure shown is a prediction result diagram of the DS-BP neural network test set in one embodiment of the present invention. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0018] Example 1 like Figure 1 As shown in the figure, this invention provides a fault detection method for a dual-bus system, comprising the following steps: S1: Obtain the branch current and bus tie current on each of the two buses in a dual-bus system; S2: Solve for the large differential current and braking current under the double bus system. If the ratio of the large differential current to the braking current is greater than or equal to the ratio braking coefficient, it is determined that the fault occurs within the area and the double bus system fault detection step is entered. Otherwise, it is determined that the fault occurs outside the area and the protection does not operate. S3: When the fault is determined to occur within the region, calculate the small differential current of the two buses in the dual-bus system, process the large differential current and the small differential current of the two buses to obtain time-domain characteristic data. a 1- a 10 ; S4: Based on the DS evidence theory, the time-domain feature data of the three currents are fused to obtain the time-domain feature training data of the dual bus system. The BP neural network is trained through the fused dataset to obtain the trained dual bus fault detection model. S5: Use this model to analyze and diagnose the time-domain characteristics of the dual busbars, thereby determining the location and type of the fault.

[0019] This embodiment provides a fault detection method for a dual-bus system. It can acquire the branch current and bus tie current of each of the two buses in the area under test and calculate the large differential current, small differential current and braking current of the bus. First, it determines whether the fault is outside the area or inside the area. When the fault occurs within the area, it enters the subsequent dual-bus system fault detection process. The dual-bus fault detection uses a trained neural network algorithm model to detect the fault type and fault location, thereby obtaining the corresponding detection results and improving the accuracy and reliability of fault detection.

[0020] In this embodiment, in step S1, the branch current and bus tie current of each of the two busbars in the area to be tested are acquired in real time through the sensors built into the microcomputer protection device. The microcomputer protection device is a key device in modern power systems. It combines computer technology, signal processing technology and automatic control technology and has the characteristics of high precision, high reliability and intelligence. Since this device is existing technology and is not the innovation of this invention, it will not be described in detail here.

[0021] In this embodiment, the method for solving the large differential current, small differential current, and braking current in step S2 includes: Double busbar large differential current To consider the double busbars and all their connected branches (including bus couplers) as a whole, the vector sum of the currents in all branches, and the differential current of the busbars. and The vector sum of the branch currents of busbar 1 and busbar 2, and the braking current. I res This is the scalar sum of all branch currents and bus tie currents.

[0022] In this embodiment, in step S2, if the ratio of the large differential current to the braking current of the busbar in the test area is greater than or equal to the ratio braking coefficient, it is determined that the fault occurs within the area, and the process proceeds to the small differential current detection step. Otherwise, it is determined that the fault occurs outside the area, and the protection does not operate. Specifically, this includes: ; in, k set The braking coefficient is usually set to 0.3~0.7, and is set to 0.3 here. When the condition is met, it is judged to be a fault in the area and proceeds to the subsequent double busbar system fault detection steps S3-S5. Otherwise, it is considered that the fault has not occurred or has occurred outside the area.

[0023] In this embodiment, the training data collection and diagnosis process of the dual-bus system fault detection model in steps S3-S5 are as follows: After obtaining the large differential current and the small differential current of the two buses in the dual-bus system, their fundamental time-domain characteristic parameters are determined. a 1- a 7 and auxiliary time-domain feature parameters a 8- a 10 To obtain; The time-domain characteristic parameters of the three currents are fused using the DS evidence theory to obtain the training data matrix X of the dual-bus fault detection model; The dual-bus fault detection model is constructed primarily using a BP neural network. After training and adjustment, it determines the fault type and location.

[0024] Specifically, time-domain characteristic parameter calculations are performed on the large differential current of the busbar and the small differential current of the two buses calculated in steps S2 and S3, including: ; Where represents the number of data points, and represents the waveform. a 1- a 7 represents key time-domain features such as mean and standard deviation, which can reflect the overall strength and volatility of the bus waveform. a 8 This represents the maximum value of the differential current waveform amplitude at the busbar. a 9 This represents the minimum amplitude of the differential current waveform at the busbar. a 10 This indicates the peak value.

[0025] The dual-bus fault detection model, based on key time-domain features such as mean and standard deviation, also records the maximum, minimum, and peak-to-peak values, providing information on different aspects of the amplitude range of the large-differential and small-differential differential current waveforms of the bus, thus assisting in the analysis of fault categories.

[0026] Example 2 Based on the fault detection method for a dual-bus system provided in Example 1, this example describes the process of optimizing the training set of the DS system, as follows: The dual-bus fault detection model employs Dempster evidence theory to optimize network input during training, comprehensively reflecting the time-domain characteristics of fault waveforms in a dual-bus system. This effectively shortens training time and reduces uncertainty. Multi-feature fusion is implemented based on a generalized form of the Dempster synthesis rule. ; in, m i For the first i The basic probability assignment function for each source of evidence. n This represents the total number of time-domain feature parameters, which is 30 in this case. A To determine the type of fault, A i The fault type supported by local evidence provided for a single feature, where K is the conflict quantity. The feature matrix used for model training is obtained in this way; the feature matrix X of a single input is [ m 1, m 2, m 3… m 10 ] T ,in m These are the fused feature parameters. Using this method, the time-domain feature parameters of the three current signals obtained in Example 1 are fused, reducing 30 input data points to 10 features and optimizing the model training effect.

[0027] After completing the training set data fusion, a backpropagation (BP) neural network was established to train a fault detection model for a dual-bus system. Multiple artificial neurons with simple functions are interconnected through a specific topology to form a neural network. The basic structure of the neurons is shown below: ; in, For the first l Layer j Net input to each neuron, Indicates the first l -1st floor i The first neuron to the second l Layer jThe connection weights of each neuron Indicates the first l -1st floor i The output of each neuron Indicates the first l Layer j The bias term of each neuron. n Indicates the first l The number of neurons in layer -1. (Through the transfer function) f(x) This invention simulates the transmission characteristics of biological neurons to achieve information transmission in artificial neurons. The BP neural network used in this invention calculates the output of each neuron backward using the known training samples inserted in the first layer. The last layer calculates the weights and thresholds forward by inputting the established network structure, the weights and thresholds of the previous iteration. In the process of continuous establishment and correction, the relationship between fault feature values ​​and fault types is established, thereby realizing bus fault diagnosis.

[0028] Meanwhile, for common feature parameters, the BP neural network uses gradient search to gradually reduce the error during network training, and chooses the continuous derivative sigmoid function as the transfer function: ; in f(x) For the transfer function, exp( x ) is an exponential function, x This is the input to the neuron.

[0029] For fault characteristics that vary in the positive and negative ranges, the transfer function of the neuron during training is a hyperbolic function, specifically a symmetric sigmoid function: ; Where is the transfer function, exp( x ) is an exponential function, x This is the input to the neuron.

[0030] After optimizing the network input of the BP neural network model based on DS evidence theory, this embodiment also optimizes the number of hidden layer neurons in the neural network model. By adopting trial and error training, the number of neurons was finally set to 80, which ensured training accuracy, shortened training time, and avoided network prediction errors caused by too few hidden layer nodes and overfitting problems caused by too many nodes.

[0031] In addition, when choosing the learning rate of the network η When, if the selected η If the value is too small, it will prolong the network's training time and slow down the convergence speed. ηThe value is too large. During the weight adjustment process, the changes in network connection weights will be too large. To ensure the stability of network model training, this embodiment selects 0.1 as the learning rate for the fault diagnosis model of the dual-bus system.

[0032] This invention optimizes the number of hidden layer neurons in a BP neural network-based fault diagnosis network model through stepwise experiments using empirical formulas. Too few hidden layer nodes increase network prediction error, training time, and fail to accurately represent the correspondence between network input and output; too many nodes lead to overfitting, causing the network to over-adapt to noise and errors in the data, thus hindering generalization to new data. To shorten training time and ensure training accuracy, the number of neurons is set to 80. Considering the stability of the training process, the maximum training time for the bus fault neural network is set to 1000, and the learning rate of the grid is [not specified]. η It is 0.1.

[0033] Example 3 This embodiment establishes a simulation model for a double-busbar structure, and its power system schematic diagram is shown below. Figure 2 As shown. The system voltage level is 500kV, and the specific system parameters are as follows: System impedance: ZEM1 = ZEM2 = ZEM3 = ZEM4 = 5.74 + j14.18; Transmission line parameters: R1=0.021 Ω / km, R0=0.021 Ω / km; L1=0.898 mH / km, L2=2.289 mH / km; C1=0.013 uF / km, C2=0.005 uF / km; Based on this, this embodiment uses MATLAB's SIMULINK as the main platform for model building, and establishes a primary system simulation model under a dual-bus system, such as... Figure 3 As shown, by using a current transformer, the current waveform data of the large differential current and the small differential current of the busbar when a fault occurs are obtained.

[0034] The busbar full current differential system mainly consists of fault starting elements, large differential differential elements, I bus small differential differential elements, II bus small differential differential elements, TA saturation identification elements, and TA disconnection detection elements. The starting elements are composed of the QD xinhao subsystem, the TA saturation elements are composed of the TA saturation subsystem, the large differential differential elements are composed of the large differential differential subsystem, and the I and II bus small differential differential elements are composed of the small differential differential subsystem. The system simulation time is set to 0-0.2 s, and the simulation algorithm selected is ode23t.

[0035] Example 4 This invention simulates faults within a dual-busbar region to obtain waveform data of the large-differential busbar differential current and the small-differential busbar differential current when a fault occurs. The simulateable faults are shown in Table 1. Table 1. Simulation Fault Types of Dual Busbar Systems ; This example simulates a three-phase short-circuit fault on a single busbar based on the dual-busbar simulation model established in Example 3. Figure 4 The diagram shows the effective values ​​of the large differential current and the small differential current of the busbar when a three-phase short-circuit fault occurs. Figure 5 This is a graph showing the changes in the operating signals of large-differential and small-differential components. From... Figure 5 As can be seen, when a three-phase short-circuit fault occurs on a busbar, the large differential element and the small differential element of the first busbar operate at 0.105s, as their operating signals change from 0 to 1. Meanwhile, the small differential elements of the second busbar do not operate, as their operating signals remain at 0. This indicates that the fault occurred on a busbar within the area, verifying the accuracy of the fault location simulation. By comparing the effective waveforms of the small differential currents of the first and second buses, it can be observed that when the fault occurs, the waveform and amplitude of the small differential current of the first busbar exhibit a regular abrupt change compared to the second busbar. This provides the basis for introducing the DS-BPNN neural network diagnostic model of this invention. By performing time-domain analysis on the fault current effective waveform, time-domain feature parameters that effectively reflect the characteristics of the busbar fault are extracted, thereby enabling accurate diagnosis of the busbar fault using a neural network algorithm model.

[0036] Example 5 After establishing the fault detection model for the dual-bus system constructed in Example 2, this example uses the mean square error (MSE) to characterize the training results in detail, as shown in the following figures. Figure 6 As shown, Mean Squared Error (MSE) is a commonly used metric to evaluate the difference between model predictions and actual observations. A smaller MSE value indicates a smaller discrepancy between the model's predictions and the true values, and a better model fit. The MSE curve clearly shows that as the training epochs increase, the MSE steadily decreases, gradually approaching the set target value. When the 80th epoch is reached, the training error has reached the expected level, and training ends prematurely.

[0037] In this embodiment, changes in training gradients and training errors are as follows: Figure 7As shown, the training gradient decreases rapidly with increasing training cycles, indicating that the model converges smoothly and reaches the optimal solution during training. Furthermore, the number of training errors is virtually zero, demonstrating the model's excellent performance and stability during the training phase.

[0038] Example 6 In this example, based on the fault sample data obtained from the bus dynamic simulation, each set of data consists of waveform data measured from the bus large differential current transformer and the bus small differential current transformer in a dual bus system. By performing time-domain analysis on the fault data, time-domain characteristic parameters are obtained, and then evidence fusion is performed to obtain the input of the network model.

[0039] This example compares and validates the test dataset by feeding it into PCA-ANN, PSO-SOM, DNN, and DS-BPNN networks respectively. The detection accuracy results are shown in Table 2. Table 2. Detection accuracy of different neural network fault detection models ; As shown in the table above, the detection accuracy of DS-BPNN reached 100%, which is higher than that of PCA-ANN, PSO-SOM, and DNN networks. The comparison between its prediction results and actual results is shown in the figure below. Figure 8 As shown, the predicted values ​​fit the true values ​​well. The DS-BPNN neural network, namely the fault detection model based on a dual-bus system proposed in this invention, effectively improves the accuracy of fault detection in dual-bus systems.

[0040] Example 7 This embodiment provides a fault detection system for a dual-bus system, including: The data acquisition module is used to acquire the branch current and bus tie current on each of the two buses in a dual-bus system. The judgment module is used to solve the large differential current and braking current in a double bus system. If the ratio of the large differential current to the braking current is greater than or equal to the ratio braking coefficient, it is determined that the fault occurs within the area and the double bus system fault detection step is entered. Otherwise, it is determined that the fault occurs outside the area and the protection does not operate. The time-domain module is used to solve the small differential current of the two buses in a dual-bus system when the fault is determined to occur within the region. It processes the large differential current and the small differential current of the two buses to obtain time-domain characteristic data. The training module is used to fuse the time-domain feature data of the large differential current and the small differential current of the two buses based on the DS evidence theory to obtain the time-domain feature training data of the dual-bus system. The BP neural network is then trained using the fused dataset to obtain the trained dual-bus fault detection model. The detection module is used to analyze and diagnose the time-domain characteristics of the dual busbars using a trained dual busbar fault detection model, thereby determining the location and type of the fault.

[0041] In summary, this invention, based on the principle of bus differential protection, establishes an intelligent diagnostic model for bus faults. First, the principle of traditional bus differential protection is analyzed, and a specific analysis of faults in a double bus system is conducted. Based on this, data from the current transformers on each branch and bus tie in the double bus system are collected, and the large and small differential currents of the bus are calculated. Time-domain analysis of the fault waveform is performed to obtain time-domain characteristic parameters, which effectively reflect the characteristics of bus faults. To further improve training accuracy and speed, evidence fusion of the time-domain characteristic parameters is performed using DS evidence theory, transforming the time-domain characteristic parameters of the three current signals into joint characteristic parameters for neural network training. A BP neural network is selected for training, trained based on fault samples obtained from bus dynamic simulation. Based on error backpropagation, 80 hidden layers are established to establish the connection between bus characteristic values ​​and bus faults. In verifying the model's accuracy, the diagnostic accuracy rate is 100% when analyzing test fault data, indicating that the model can achieve accurate diagnosis of bus faults.

[0042] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A fault detection method for a dual-bus system, characterized in that, include: Obtain the branch current and bus tie current on each of the two buses in a dual-bus system; Solve for the large differential current and braking current in a double bus system. If the ratio of the large differential current to the braking current is greater than or equal to the ratio braking coefficient, it is determined that the fault occurs within the area and the double bus system fault detection step is initiated. Otherwise, it is determined that the fault occurs outside the area and the protection does not operate. When the fault is determined to occur within the area, the small differential current of the two buses in the dual-bus system is calculated. The large differential current and the small differential current of the two buses are processed to obtain time-domain characteristic data. Based on the DS evidence theory, the time-domain characteristic data of the large differential current and the small differential current of the two buses are fused to obtain the time-domain characteristic training data of the dual-bus system. The BP neural network is trained through the fused dataset to obtain the trained dual-bus fault detection model. The time-domain characteristics of the dual busbars are analyzed and diagnosed using a trained dual busbar fault detection model to determine the location and type of the fault.

2. The fault detection method for a dual-bus system according to claim 1, characterized in that, The methods for solving the large differential current, small differential current, and braking current include: Double busbar large differential current The vector sum of the currents in all branches after treating the two busbars and all their connected branches as a whole; Busbar differential current and This is the vector sum of the branch currents of busbar 1 and busbar 2; Braking current I res This is the scalar sum of all branch currents and bus tie currents.

3. The fault detection method for a dual-bus system according to claim 2, characterized in that, Fault identification within and outside the area includes: ； in, k set The braking coefficient is used to determine whether a fault occurs within the area when the conditions are met, and to proceed with the subsequent double busbar fault detection process. Otherwise, it is assumed that the fault has not occurred or has occurred outside the area.

4. The fault detection method for a dual-bus system according to claim 3, characterized in that, The training data collection and diagnostic process for the fault detection model of the dual-bus system includes: After obtaining the large differential current and the small differential current of the two buses in the dual-bus system, their fundamental time-domain characteristic parameters are determined. a 1- a 7 and auxiliary time-domain feature parameters a 8- a 10 To obtain; The time-domain characteristic parameters of the three currents are fused using the DS evidence theory to obtain the training data matrix X of the dual-bus fault detection model; The dual-bus fault detection model is constructed primarily using a BP neural network. After training and adjustment, it determines the fault type and location.

5. A fault detection method for a dual-bus system according to claim 4, characterized in that, The time-domain fundamental characteristic parameters of the large differential current of the busbar and the small differential currents of the two buses are calculated as follows: ； In the formula, N Indicates the number of data points. x(n) Indicates waveform.

6. A fault detection method for a dual-bus system according to claim 4, characterized in that, The auxiliary time-domain feature parameters are calculated as follows: ； In the formula, a 8 This represents the maximum value of the differential current waveform amplitude at the busbar. a 9 This represents the minimum amplitude of the differential current waveform at the busbar. a 10 This indicates the peak value.

7. A fault detection method for a dual-bus system according to claim 4, characterized in that, The dual-bus fault detection model employs Dempster evidence theory to optimize network input during training. Multi-feature fusion is implemented based on a generalized form of Dempster's synthesis rules. ； In the formula, m i For the first i The basic probability assignment function for each source of evidence, where n is the number of all time-domain feature parameters. A To determine the type of fault, A i The type of failure supported by local evidence provided for a single feature, where K is the conflict quantity.

8. A fault detection method for a dual-bus system according to claim 4, characterized in that, The dual-bus fault detection model is trained using a BP neural network, which iteratively performs forward and backward propagation. The BP neural network gradually optimizes the weights and biases, and finally fits the relationship between fault features and fault types.

9. A fault detection method for a dual-bus system according to claim 4, characterized in that, For the dataset optimized by DS evidence theory, the BP neural network uses gradient search to gradually reduce the error, and the transfer function is chosen to be the continuous derivative sigmoid function: ； In the formula, f(x) For the transfer function, exp( x ) is an exponential function, x This is the input to the neuron.

10. A fault detection system for a dual-bus system, characterized in that, include: The data acquisition module is used to acquire the branch current and bus tie current on each of the two buses in a dual-bus system. The judgment module is used to solve the large differential current and braking current in a double bus system. If the ratio of the large differential current to the braking current is greater than or equal to the ratio braking coefficient, it is determined that the fault occurs within the area and the double bus system fault detection step is entered. Otherwise, it is determined that the fault occurs outside the area and the protection does not operate. The time-domain module is used to solve the small differential current of the two buses in a dual-bus system when the fault is determined to occur within the region. It processes the large differential current and the small differential current of the two buses to obtain time-domain characteristic data. The training module is used to fuse the time-domain feature data of the large differential current and the small differential current of the two buses based on the DS evidence theory to obtain the time-domain feature training data of the dual-bus system. The BP neural network is then trained using the fused dataset to obtain the trained dual-bus fault detection model. The detection module is used to analyze and diagnose the time-domain characteristics of the dual busbars using a trained dual busbar fault detection model, thereby determining the location and type of the fault.

Images