Selective protection method for low-voltage system containing direct-current source load based on XGBoost algorithm

Abstract

The invention discloses an XGBoost algorithm-based direct-current source load-containing low-voltage system selective protection method. The method comprises the following steps of: extracting a fifth-scale detail component dd5 of a current signal in real time; comparing the value of the fifth scale detail component with a preset threshold, and only uploading threshold exceeding data; constructing a feature vector based on the uploaded data, and inputting the feature vector into an XGBoost model to locate an alternating current side fault; and the direct current side directly triggers the circuit breaker to open through local threshold comparison. An alternating current system fault position identification scheme based on an XGBoost algorithm is combined with a direct current system in-situ protection scheme based on short circuit fault early detection to form a low-voltage alternating current and direct current system selective protection method. The method is beneficial to improving the rapidity, sensitivity and selectivity of low-voltage AC / DC system protection.

Description

Technical Field

[0001] The present invention relates to the technical field of power detection, and particularly to a selective protection method for a low-voltage system with DC sources and loads based on the XGBoost algorithm. Background Art

[0002] At present, with the large-scale grid connection of distributed power sources and the rapid development of DC loads such as energy storage and charging piles, the AC-DC hybrid distribution network has become the future development trend. Under this background, the complexity of the operating conditions of low-voltage power distribution and utilization systems is increasing day by day, and traditional low-voltage short-circuit protection technologies are facing many new problems and challenges, which mainly stem from the coupling and dynamic interaction of faults on both the AC and DC sides.

[0003] Compared with traditional low-voltage AC systems, the topological structure and source-load characteristics of low-voltage AC-DC systems have undergone fundamental changes. On the one hand, current research on AC-DC hybrid systems mainly focuses on dispatching, planning, and control strategies, and the research on the protection of low-voltage AC-DC systems is relatively scarce. On the other hand, the operating conditions of low-voltage multi-level AC-DC systems are diverse, and there is dynamic interaction and coupling between the fault currents on both the AC and DC sides. The fault currents of different levels-branches exhibit the characteristics of electric network transmission.

[0004] It is worth noting that the source and load access conditions of the DC system have a significant impact on the AC-side fault characteristics of the low-voltage AC-DC system. Especially with the popularization and application of new energy power generation such as photovoltaics, the installed capacity of the DC system continues to increase, and the penetration rate continues to rise, resulting in an increasingly greater impact on the AC fault characteristics. In addition, DC-side faults are characterized by large fault currents and short peak arrival times, and will cause overcurrent on the AC side, which poses higher requirements for the rapidity, sensitivity, and selectivity of the protection of low-voltage AC-DC systems.

[0005] Therefore, there is an urgent need to propose a low-voltage AC-DC short-circuit protection method applicable to the access of multi-condition DC systems to improve the safety and reliability of the system. Summary of the Invention

[0006] In view of the defects and deficiencies existing in the prior art, the present invention provides a selective protection method and system for a low-voltage system with DC sources and loads based on the XGBoost algorithm, and its innovative design points include:

[0007] Hierarchical protection architecture: Through the cooperation of the distributed fault detection module (FDM) and the central control module (CCM), accurate positioning on the AC side and fast local protection on the DC side are realized. The FDM locally performs three-order B-spline wavelet packet decomposition, extracts the fifth-scale detail component (dd5) and screens out the data exceeding the threshold. The CCM centrally constructs a sparse feature vector and inputs it into the XGBoost model for fault location, reducing communication delay and data redundancy.

[0008] Fault feature extraction and screening mechanism: The fifth scale detail component (DD5) is innovatively used as the singular characteristic value of the short-circuit fault, combined with the communication window of 20 sampling points and the time difference limit of 4 sampling points to ensure the timing synchronization of fault data and the reliability of decision-making.

[0009] Differentiated protection strategies:

[0010] AC side: The XGBoost model is used to fuse the dd5 peak data of multiple detection points and the fault type identifier (0-6), and the probability matrix is output to achieve high-precision positioning of the fault branch;

[0011] DC side: Local threshold setting and direct tripping mechanism to avoid protection misoperation or refusal to operate due to AC-DC fault coupling.

[0012] Threshold optimization and model training: The fixed threshold on the DC side is adjusted based on experiments, and the XGBoost hyperparameters (tree depth, learning rate, regularization coefficient) are optimized through cross-validation to improve the generalization ability of the model.

[0013] The present invention solves the problems of insufficient rapidity, sensitivity and selectivity of the traditional method in the AC / DC hybrid system, and significantly improves the safety and reliability of the low-voltage AC / DC system under complex working conditions.

[0014] The present invention specifically adopts the following technical solutions:

[0015] A selective protection method for a low-voltage system with DC source and load based on an XGBoost algorithm, comprising:

[0016] Extract the fifth scale detail component dd5 of the current signal in real time;

[0017] Comparing the value of the fifth scale detail component with a preset threshold, and uploading only data exceeding the threshold;

[0018] Build feature vectors based on uploaded data and input them into the XGBoost model to locate AC side faults;

[0019] The DC side directly triggers the circuit breaker to open through local threshold comparison.

[0020] Furthermore, the fifth scale detail component is extracted by performing 5-layer wavelet packet decomposition using a cubic B-spline wavelet basis function.

[0021] Furthermore, the feature vector includes dd5 peak data and fault type identifiers of multiple detection points, and the corresponding components of the detection points that do not exceed the threshold are 0.

[0022] Furthermore, the receiving window length of the uploaded data is 20 sampling points, and the triggering time difference of the fault branch data does not exceed 4 sampling points.

[0023] Furthermore, the preset threshold value on the DC side is a fixed value determined by experimental tuning.

[0024] Furthermore, the value range of the fault type identifier is from 0 to 6, corresponding to different fault scenarios.

[0025] A selective protection system for a low-voltage system with DC sources and loads based on the XGBoost algorithm includes:

[0026] Multiple fault detection modules: Deployed at each detection point on the AC side, used to collect current signals in real time, extract the fifth-scale detail component dd5 through cubic B-spline wavelet packet decomposition, and only upload the data exceeding the threshold after comparing the dd5 value with the preset threshold;

[0027] A central control module: Communicatively connected to all the fault detection modules, receives the data exceeding the threshold and constructs a feature vector, inputs the XGBoost model to locate the fault on the AC side, and sends a tripping command to the corresponding circuit breaker;

[0028] A local protection mechanism on the DC side: When any fault detection module detects that the dd5 value on the DC side exceeds the threshold, directly trigger the local circuit breaker to trip.

[0029] Furthermore, the fifth-scale detail component is extracted through 5-layer wavelet packet decomposition using the cubic B-spline wavelet basis function.

[0030] Furthermore, the feature vector includes the dd5 peak data of multiple detection points and the fault type identifier, and the corresponding components of the detection points that do not exceed the threshold are 0.

[0031] Furthermore, the receiving window length for the fault detection module to upload data to the central control module is 20 sampling points, and the triggering time difference of the fault branch data does not exceed 4 sampling points.

[0032] And, an electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0033] A non-transitory computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0034] Compared with the prior art, the present invention and its preferred solutions at least include the following beneficial effects:

[0035] Improve the speed and accuracy of fault location: By combining the hierarchical protection architecture (FDM+CCM cooperation mechanism), the centralized decision-making on the AC side and the local protection on the DC side are combined, solving the problem of protection delay caused by AC-DC fault coupling and significantly shortening the fault response time.

[0036] Enhance the reliability of fault feature extraction: Based on the fifth-scale detail component (dd5) extracted by cubic B-spline wavelet packet decomposition, it can effectively capture the mutation characteristics of short-circuit current. Combining with the threshold screening mechanism can reduce redundant data transmission and reduce the impact of noise interference on model decision-making.

[0037] Optimize the protection selectivity under complex working conditions: The differential protection strategy (XGBoost probability location on the AC side and direct tripping by threshold on the DC side) takes into account the requirements of multi-branch fault identification on the AC side and rapid isolation on the DC side, avoiding misoperation and refusal to operate of traditional overcurrent protection in AC-DC hybrid systems.

[0038] Improve the model generalization ability and system adaptability: Through feature vector sparsification (assigning 0 to data not exceeding the threshold) and fault type identifier classification, the robustness of the XGBoost model in multi-detection points and multi-fault scenarios is strengthened to adapt to the dynamic changes of the low-voltage AC-DC system topology.

[0039] Simplify the implementation complexity and resource consumption: The combination of local processing of the distributed detection module (FDM) and the communication window mechanism greatly reduces the computing burden of the central control module (CCM), and at the same time reduces the communication bandwidth requirement, which is applicable to low-voltage distribution scenarios with limited resources.

[0040] The present invention forms a complete technical closed-loop from the method, architecture to the implementation level, providing a protection solution for low-voltage AC-DC systems that takes into account speed, sensitivity and selectivity. Brief Description of the Drawings

[0041] The following further describes the present invention in detail with reference to the drawings and specific embodiments:

[0042] Figure 1 It is the wavelet packet decomposition structure diagram in the embodiment of the present invention;

[0043] Figure 2 It is the XGBoost model training process diagram in the embodiment of the present invention;

[0044] Figure 3 It is the AC side protection system architecture diagram in the embodiment of the present invention;

[0045] Figure 4 It is the AC side protection flow chart in the embodiment of the present invention;

[0046] Figure 5 It is the schematic diagram of the dd5 detection triggering the CCM communication window in the embodiment of the present invention. Detailed implementation manners

[0047] In the following, specific embodiments of the present application will be described in detail with reference to the accompanying drawings. According to these detailed descriptions, those skilled in the art can clearly understand the present application and can implement the present application. Without departing from the principle of the present application, the features in different embodiments can be combined to obtain new implementation manners, or some features in certain embodiments can be replaced to obtain other preferred implementation manners.

[0048] To make the features and advantages of the present invention more obvious and understandable, specific embodiments are given below and will be described in detail in conjunction with the accompanying drawings as follows:

[0049] An AC-DC system protection method based on the XGBoost algorithm is provided in an embodiment of the present invention. The artificial intelligence algorithm is used to more efficiently extract the correlation information of fuzzy data. Through learning, training and optimization of a large amount of data, a more accurate and refined decision-making effect is provided. For the complex situation of AC-side faults, the present invention combines an AC system fault location identification scheme based on the XGBoost algorithm with a DC system in-situ protection scheme based on early short-circuit fault detection to form a selective protection method for low-voltage AC-DC systems. This method is beneficial to improving the rapidity, sensitivity and selectivity of low-voltage AC-DC system protection.

[0050] As a specific implementation of the embodiment of the present invention, the current mutation information at the moment of short-circuit fault occurrence is extracted through three-order B-spline wavelet packet detail decomposition. The wavelet packet decomposition decomposes the original data and forms a complete wavelet packet tree, where U j,n is the subspace of the j-th scale wavelet packet of the n-th (where n is the frequency factor, n = 0, 1, 2, ···, 2j - 1), is its corresponding orthogonal basis, where (k is the offset factor), and its calculation formula is:

[0051]

[0052] where j, k ∈ Z, n = 0, 1, 2,..., 2 j -1, h k and g k are the low-pass and high-pass filter coefficients respectively.

[0053] Performing five-layer three-order B-spline wavelet packet detail decomposition on the input signal, we can obtain:

[0054]

[0055] In the formula, is the five-layer low-frequency component obtained by the signal passing through the low-pass filter h(n) This low-frequency component is called the approximation coefficient; It is filtered by the high-pass filter g(n) into 5 layers of high-frequency components, and this high-frequency component is called the detail coefficient.

[0056] In view of the fact that when a short-circuit fault occurs in the low-voltage AC-DC hybrid system, the singular value of the fifth scale is the most obvious after the short-circuit current is decomposed by wavelet packet. Therefore, as Figure 1 shown, in this embodiment, the detail component of the fifth scale of wavelet packet decomposition is used as the singular eigenvalue of the short-circuit fault. For the sake of convenience of expression, the detail component of the fifth scale obtained by wavelet decomposition of the original signal will be represented as dd5 in the following.

[0057] Specifically, the XGBoost algorithm consists of K base learners to form an additive operation formula as:

[0058]

[0059] In the formula, x i is the i-th input data, is the predicted value of the i-th sample, K is the number of regression trees, and f k is the k-th base model, and F is the set corresponding to all trees.

[0060] To evaluate the performance of the fault location algorithm trained by the XGBoost algorithm in different fault conditions in the low-voltage AC-DC hybrid system, the prediction accuracy rate (successrate) is used as the evaluation index, and its definition is:

[0061]

[0062] Among them, N r represents the number of samples in which the algorithm model correctly identifies and locates faults during use, then N all represents the comprehensive number of all participating test set samples.

[0063] Specifically, the peak value of dd5 of the wavelet packet detail decomposition algorithm has the position information of the short-circuit fault. By constructing the non-linear relationship between the dd5 peak values of different detection points through the artificial intelligence algorithm, the accurate identification of the fault position is realized. The sample feature vector used to construct the XGBoost model can be expressed as:

[0064] X = [dd51, dd52, dd53, …, dd58, I] I ∈ (0, 1, 2, …, 6)(5)

[0065] In the formula, dd5 x (x = 0, 1, 2, ……, 8) represents the dd5 peak value data corresponding to the detection point after threshold screening. Only the detection points with dd5 values exceeding the detection threshold can upload the dd5 data, and the corresponding dd5 xDetection points that are not zero but below the detection threshold, corresponding to dd5 x Is always zero, and the output value corresponds one-to-one with the fault identification result of the algorithm.

[0066] As a preferred solution of this embodiment, fault data is obtained through simulation means, the wavelet packet detail decomposition algorithm is run to extract the dd5 value, the dd5 output values of each detection point are compared with the set threshold, the local dd5 data threshold screening is completed, and a sample feature vector data set is generated. The simulated real-time waveform data set is converted into a numerical sample data set. The processed numerical sample data is divided into a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%. The training set is substituted into the XGBoost algorithm, and the optimal hyperparameter combination is selected through cross-validation to train the target model. Finally, the test set samples are substituted into the XGBoost model, and the test results and misjudged samples in the test set are analyzed to complete the model training work. The process of XGBoost model training is as Figure 2 shown.

[0067] As a preferred solution of this embodiment, the AC side protection system framework is as Figure 3 shown. Each fault detection point is equipped with a short-circuit fault detection module (fault detection module, FDM). Each FDM incorporates a short-circuit fault early detection algorithm and the set threshold of the corresponding location, performs wavelet packet decomposition algorithm on the current data sampled at the detection point to generate real-time dd5 data, and compares the dd5 value with the fault detection threshold. All the upper layers of the FDM modules are uniformly connected to a central control module (central control module, CCM), which is used to receive the communication data of the subordinate FDM modules, execute the AC side short-circuit fault location identification algorithm based on XGBoost, and simultaneously have the functions of fault protection decision-making and sending disconnection instructions. The current mutation characteristics of DC short-circuit faults have little correlation with the occurrence time of short-circuit faults and the steady-state operating conditions of the system, and the DC steady-state current fluctuates little. By using the wavelet packet detail decomposition algorithm and setting appropriate fault detection thresholds, the in-situ rapid identification and protection of DC side short-circuit faults can be achieved without additional backup decision-making algorithms to determine DC side feeder short-circuit faults.

[0068] As Figure 4 shown, the implementation steps of the selective protection method for the AC-DC system in this embodiment are as follows:

[0069] (a) The detection algorithm based on the cubic B-spline wavelet packet detail decomposition monitors the line current in real time.

[0070] (b) When dd5 exceeds the set threshold, upload the peak value of the dd5 component, retrieve the corresponding XGBoost model, output the probability matrix, and locate the fault point.

[0071] (c) Send a trip signal to the circuit breaker corresponding to the fault location.

[0072] As Figure 5 shown, in this embodiment, taking the fault occurrence on branch l1 as an example, the dd5 value of the FDM3 module reaches the threshold first and sends the current real-time dd5 value to the CCM module. After receiving the data, the CCM module starts the decision-making. Taking the first dd5 value sent by the FDM3 module as the starting point, it continues to receive the number of points within a certain window time. Within this window length, the CCM can receive and store the dd5 data uploaded by all subordinate FDMs exceeding the threshold. The time difference of the detection points dd5 of the faulty line reaching the threshold does not exceed four sampling points. The FDM1 will reach the threshold and send the dd5 data to the CCM shortly after the CCM opens the data reception window. The time for the dd5 value of the FDMy (y = 2, 4, 7, 8) that may exceed the threshold in the non-faulty branch to reach a peak value is also within the set window length. Considering the influence of the detection time and the reliability of fault identification, the length of the CCM data reception window is set to 20 sampling points in this paper.

[0073] After the data reception window time ends, the CCM selects the peak value in the dd5 value data of each FDM as the input feature vector value of XGBoost. For the FDM module that does not trigger the threshold within the window time, the CCM sets the input feature vector value at this position to 0, sorts out the complete feature vector and inputs it into the XGBoost model for fault identification. If the output result is zero, it indicates normal operation and no fault occurs, and continues to monitor the information at the next moment; if the output number is not zero, it is determined that a line has a fault.

[0074] Based on the same inventive concept, the present invention further provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is used to implement one or more instructions. Specifically, it is used to load and execute one or more instructions in the computer storage medium to implement the above method.

[0075] It should be further noted that, based on the same inventive concept, the present invention further provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a Random Access Memory (RAM), a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.

[0076] In the description of this specification, the descriptions referring to the terms "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0077] The above has shown and described the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure will have various changes and improvements, and these changes and improvements all fall within the scope of the present disclosure claimed.

[0078] The present invention is not limited to the above optimal implementation manner. Anyone can derive other various forms of a selective protection method for a low-voltage system with DC source and load based on the XGBoost algorithm under the inspiration of the present invention. All equal changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope covered by the present invention.

Claims

1. A selective protection method for a low-voltage system with DC source and load based on the XGBoost algorithm, characterized in that: The fifth-scale detail component dd5 of the current signal is extracted in real time; The value of the fifth-scale detail component is compared with a preset threshold, and only the data exceeding the threshold is uploaded; Based on the uploaded data, a feature vector is constructed and input into the XGBoost model to locate the AC-side fault; On the DC side, the circuit breaker is directly tripped by local threshold comparison.

2. The selective protection method for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 1, characterized in that: The fifth-scale detail component is extracted by performing 5-layer wavelet packet decomposition using the cubic B-spline wavelet basis function.

3. A selective protection method for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 1, characterized in that: The feature vector includes the dd5 peak data of multiple detection points and the fault type identifier, and the corresponding components of the detection points that do not exceed the threshold are 0.

4. A selective protection method for a low-voltage system with DC sources and loads based on the XGBoost algorithm according to claim 1, characterized in that: The receiving window length of the uploaded data is 20 sampling points, and the triggering time difference of the fault branch data does not exceed 4 sampling points.

5. A selective protection method for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 1, characterized in that: The preset threshold on the DC side is a fixed value determined by experimental tuning.

6. A selective protection method for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 3, characterized in that: The value range of the fault type identifier is from 0 to 6, corresponding to different fault scenarios.

7. A selective protection system for a low-voltage system with DC sources and loads based on the XGBoost algorithm, characterized in that, It includes: Multiple fault detection modules: Deployed at each detection point on the AC side, used to collect current signals in real time, extract the fifth-scale detail component dd5 through cubic B-spline wavelet packet decomposition, and only upload the data exceeding the threshold after comparing the dd5 value with the preset threshold; Central control module: Communicatively connected to all the fault detection modules, receives the data exceeding the threshold and constructs a feature vector, inputs it into the XGBoost model to locate the AC-side fault, and sends a tripping instruction to the corresponding circuit breaker; DC-side local protection mechanism: When any fault detection module detects that the dd5 value on the DC side exceeds the threshold, the local circuit breaker is directly tripped.

8. The selective protection system for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 7, characterized in that: The fifth-scale detail component is extracted by performing 5-layer wavelet packet decomposition using the cubic B-spline wavelet basis function.

9. The selective protection system for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 7, characterized in that: The feature vector includes the dd5 peak data of multiple detection points and the fault type identifier, and the corresponding components of the detection points that do not exceed the threshold are 0.

10. A selective protection system for a low-voltage system with DC source and load based on the XGBoost algorithm according to claim 7, characterized in that: The receiving window length for the fault detection module to upload data to the central control module is 20 sampling points, and the triggering time difference of the fault branch data does not exceed 4 sampling points.

Images