A distribution cabinet for detecting faults and cutting off faulty lines based on big data

Abstract

The present invention relates to a power distribution cabinet for detecting faults and cutting off faulty lines based on big data, and belongs to the technical field of power distribution detection. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data comprises a signal acquisition module, a signal decomposition module, a model training module and a fault detection module; the signal acquisition module is used to collect the current time series signals of each section line in real time, and mark the line section number where the current time series signal occurs; the signal decomposition module is used to perform signal decomposition processing on the collected current time series signal using a discrete wavelet transform method to extract the fault feature vector; the model training module is used to construct an ANN model cluster. The disjoint self-help aggregation technology is used to divide multiple disjoint subsets to train independent ANN models, and then aggregate the classification results of the ANN cluster; thereby reducing errors, alleviating the deviation of a single model, and improving the accuracy and reliability of the overall model cluster.

Description

Technical Field

[0001] The present invention belongs to the technical field of power distribution detection, and in particular, relates to a power distribution cabinet for detecting faults based on big data and cutting off faulty lines. Background Art

[0002] The distribution network has a complex structure and numerous branch lines. In addition, the neutral point of the distribution network line is mostly ungrounded or grounded through an arc suppression coil. This makes it difficult to determine the faulty line and branch when a distribution network line fails. The power supply department needs to deploy a large number of line patrol and troubleshooting personnel, which increases the maintenance and operation costs of the power supply company.

[0003] At present, the detection of distribution network line faults mainly relies on the small current grounding system analysis software and line fault indicators of the substation: the small current grounding system analysis software uses the change of zero-sequence current or zero-sequence voltage to determine whether a grounding fault has occurred and indicate the approximate fault line. However, this method can only determine a single type of fault and cannot accurately locate the fault line and section.

[0004] In order to further locate the fault, it is usually necessary to cooperate with the line fault indicator: the line fault indicator increases the current of the fault line by briefly inserting a small grounding resistor, and uses the position of the indicator flip to determine the specific fault location. This method requires manual inspection and is inefficient. Therefore, more advanced technical means are urgently needed to improve the fault detection capability of the distribution network, and thus the fault detection and disconnection technology based on big data and artificial intelligence technology came into being. Summary of the invention

[0005] In order to solve the technical problem that the current software analysis method can only identify a single type of fault and cannot accurately locate the fault line, the present invention provides a distribution cabinet that detects faults based on big data and cuts off the fault line.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A power distribution cabinet for fault detection based on big data, comprising a signal acquisition module, a signal decomposition module, a model training module and a fault detection module;

[0008] The signal acquisition module is used to collect the current timing signals of each section line in real time and mark the line section number where the current timing signal occurs;

[0009] The signal decomposition module is used to perform signal decomposition processing on the collected current time series signal using the discrete wavelet transform (DWT) method to extract the fault feature vector;

[0010] Preferably, the discrete wavelet transform (DWT) method is used for signal decomposition processing, and the specific process includes:

[0011] Splitting the current time series signal into current data of multiple fixed time windows;

[0012] Use DWT to perform multi-layer decomposition on the current data, select db-4 wavelet as the wavelet basis function, and decompose the current data into low-frequency components (approximate components) and high-frequency components (detail components);

[0013] Extract the high-frequency components at each level and calculate the bi-norm of the high-frequency components. The calculation formula can be expressed as:

[0014] ;

[0015] In the formula, is the second norm of the high-frequency component; is the i-th element of the high-frequency component; n is the total number of high-frequency components;

[0016] The two-norm arrays at each level are synthesized into a fault feature vector dataset , which is used for subsequent signal analysis; its expression is:

[0017] ;

[0018] In the formula, is the bi-norm of the i-th element of the high-frequency component; n is the total number of high-frequency components.

[0019] The model training module is used to build an artificial neural network (ANN) model cluster. The specific construction process includes:

[0020] Data preparation: Collect current data of various known fault types with different line section numbers and mark the line section numbers; known fault types include short circuit faults, overload faults, and ground faults;

[0021] Data processing: Use DWT to perform multi-layer decomposition of the signal, extract the high-frequency components of each layer and calculate the two-norm; combine the two-norm groups of each layer into a fault feature vector data set;

[0022] Dataset division: The fault feature vector dataset is randomly divided into multiple disjoint subsets using the disjoint bagging technique, each of which corresponds to a line section number.

[0023] Model training: Each subset is used to train an independent ANN model, that is, each independent ANN model is labeled with the corresponding line section number;

[0024] Model integration: All trained independent ANN models are integrated together to form an ANN model cluster with line section numbers.

[0025] Preferably, the specific process of using each subset to train an independent ANN model includes:

[0026] The network structure of each ANN model includes an input layer, a hidden layer, and an output layer, and the number and type of neurons in each structural layer are determined; the fault feature vector data set consists of two dimensions, time window and measurement, which are used as neurons in the input layer respectively; the number of neurons in the hidden layer is determined by the Bayesian optimization method; the neurons in the output layer are different fault types;

[0027] Use Bayesian optimization (BO) method to adjust ANN model hyperparameters (such as learning rate, number of iterations, and number of hidden layer neurons); select appropriate activation function and loss function, and use back propagation algorithm to update ANN model parameters;

[0028] Use different subsets to cross-test the performance of each independent ANN model, adjust and optimize the model parameters; until all ANN model performance indicators reach the set convergence value.

[0029] Preferably, the fault feature vector data set also includes a voltage change rate, that is, a voltage change rate in the same time window; meanwhile, the neurons in the input layer of the ANN model also include a voltage change rate.

[0030] Preferably, the activation function is Sigmoid, Tanh or ReLU function; the loss function is mean square error (MSE) or cross entropy loss function.

[0031] Preferably, the specific process of adjusting the ANN model hyperparameters using the Bayesian optimization method includes:

[0032] a) Use the objective function to define the range of hyperparameters that need to be optimized (such as learning rate, number of hidden layers, number of hidden layer neurons, activation function, etc.); and use the Gaussian process method to model the objective function;

[0033] b) Initialization: randomly select a set of initial hyperparameter combinations and train the ANN model to obtain the corresponding performance indicators;

[0034] c) train and test the corresponding ANN model on the disjoint subsets, and evaluate the performance indicators of the model for the hyperparameter combination;

[0035] d) Estimate the new objective function range: Use the probabilistic surrogate model to estimate the new objective function range based on the existing hyperparameter values ​​and their corresponding performance indicators;

[0036] e) selecting the next hyperparameter combination using an acquisition function according to the new objective function range; the acquisition function includes an expected improvement function or a probability improvement function;

[0037] f) Repeat steps ce) until the model performance index reaches a convergence value.

[0038] Preferably, the ANN model performance indicators include accuracy, recall and F1 score; the set convergence values ​​are: accuracy ≥ 97%; recall ≥ 95%.

[0039] The fault detection module is used to deploy the trained ANN model cluster; and input the fault feature vector data set extracted by the signal decomposition module into each independent ANN model for fault detection and output the fault type;

[0040] Count the fault types output by each independent ANN model, and select the fault type with the most votes as the final detection result;

[0041] At the same time, among the independent ANN models that meet the final detection results, the line section number with the most votes is selected as the fault location.

[0042] Preferably, the fault types include short circuit fault, overload fault and ground fault.

[0043] A power distribution cabinet for detecting faults and cutting off faulty lines based on big data, also includes a fault cutting module, the fault cutting module includes a circuit breaker and a main controller, a plurality of circuit breakers are respectively arranged at the end of each line section, and are used to cut off the faulty line; the main controller is respectively communicated with the plurality of circuit breakers, and is used to control the cutting instructions of the circuit breakers of the corresponding line section according to the detection results of the fault detection module and the fault location.

[0044] Beneficial effects of the present invention:

[0045] 1. The wavelet transform method is used to extract features from the collected current time series signals to provide effective feature information for subsequent model training; the disjoint self-service aggregation (Bagging) technology is used to divide multiple disjoint subsets to train independent ANN models, and then the classification results of the ANN cluster are aggregated; thereby reducing errors, alleviating the deviation of a single model, and improving the accuracy and reliability of the overall model cluster.

[0046] 2. By defining the objective function and search range, Bayesian optimization can efficiently find the best hyperparameter combination to improve the performance of the ANN model.

[0047] 3. By reasonably setting circuit breakers at the end of each line section, the faulty line can be cut off quickly, accurately and intelligently, ensuring that the rest of the distribution system can continue to operate normally and ensuring the safe and stable operation of the distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0049] Figure 1 This is a schematic diagram of the modular structure of a power distribution cabinet for cutting off faulty lines based on big data fault detection according to the present invention.

[0050] Figure 2 It is a schematic diagram of the multi-level decomposition structure of a discrete wavelet transform (DWT) method in a distribution cabinet for fault detection based on big data according to the present invention.

[0051] Figure 3 This is a flow chart of a method for detecting faults in a power distribution cabinet based on big data according to the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] See also Figure 1-Figure 3 As shown, a power distribution cabinet for detecting faults and cutting off faulty lines based on big data includes a signal acquisition module, a signal decomposition module, a model training module and a fault detection module;

[0054] The signal acquisition module is used to collect the current timing signals of each section line in real time and mark the line section number where the current timing signal occurs;

[0055] Specifically, Advantech USB-4711A data acquisition cards are used and installed in key locations of power distribution lines and electrical equipment to achieve real-time data acquisition and data recording functions. The sampling rate of the data acquisition card determines the number of data points collected per second. The higher the sampling rate, the more detailed the data collected.

[0056] In addition, the collected signals can be processed in combination with the LabVIEW programming environment, including noise filtering; removing noise from the signal and improving data accuracy.

[0057] The signal decomposition module is used to perform signal decomposition processing on the collected current time series signal using the discrete wavelet transform (DWT) method to extract the fault feature vector;

[0058] Furthermore, the discrete wavelet transform (DWT) method is used to perform signal decomposition processing, and the specific process includes:

[0059] The current time series signal is divided into current data of multiple fixed time windows; for example, fixed time windows such as 1 second and 10 milliseconds.

[0060] Use DWT to perform multi-layer decomposition on the current data, select db-4 wavelet as the wavelet basis function, and decompose the current data into low-frequency components (approximate components) and high-frequency components (detail components);

[0061] Extract the high-frequency components at each level and calculate the second norm of the high-frequency components. The calculation formula can be expressed as: current data

[0062] ;

[0063] In the formula, is the second norm of the high-frequency component; is the i-th element of the high-frequency component; n is the total number of high-frequency components;

[0064] The two-norm arrays at each level are synthesized into a fault feature vector dataset , which is used for subsequent signal analysis; its expression is:

[0065] ;

[0066] In the formula, is the bi-norm of the i-th element of the high-frequency component; n is the total number of high-frequency components.

[0067] Specifically, wavelet transform is a signal processing method that uses wavelet functions to decompose signals into sub-signals of different frequency components; wavelet transform can effectively extract signal features and compress and reduce noise on the signal. The db-4 wavelet is an orthogonal wavelet, which means that the sub-signals of different frequency components are orthogonal to each other and will not interfere with each other. This ensures the stability and reliability of the wavelet decomposition process and avoids confusion between different frequency components.

[0068] Among them, the low-frequency component contains the main energy and trend information of the signal, and the high-frequency component contains the details and mutation information of the signal. The second norm (also known as the Euclidean norm) refers to the square root of the sum of the squares of the signal vector. It can be used to measure the energy or strength of the signal, that is, expressed as a metric. By calculating the second norm of the high-frequency component, the high-frequency component in the signal can be effectively quantified, which is helpful for fault detection and signal feature analysis.

[0069] For high-frequency components, a higher number of decomposition layers is required to extract more detailed information. The appropriate number of decomposition layers can be selected based on manual fault analysis experience or the frequency range of the signal. For example, distribution network lines usually use 50Hz-100Hz AC power, and 4-6 layers of decomposition are generally selected; 4-6 layers of decomposition can extract enough detailed information while maintaining a reasonable computational complexity, balancing performance and efficiency.

[0070] Furthermore, the extracted feature vectors can be standardized, and various data standardization methods, such as normalization and standardization, can be used to map the data range to a unified range. Data standardization can improve the efficiency of model training and avoid excessive influence of certain data on model training.

[0071] The model training module is used to build an artificial neural network (ANN) model cluster. The specific construction process includes:

[0072] Data preparation: Collect current data of various known fault types with different line section numbers and mark the line section numbers; known fault types include short circuit faults, overload faults, and ground faults;

[0073] Data processing: Use DWT to perform multi-layer decomposition of the signal, extract the high-frequency components of each layer and calculate the two-norm; combine the two-norm groups of each layer into a fault feature vector data set;

[0074] Dataset division: The fault feature vector dataset is randomly divided into multiple disjoint subsets using the disjoint bagging technique, each of which corresponds to a line section number.

[0075] Model training: Each subset is used to train an independent ANN model, that is, each independent ANN model is labeled with the corresponding line section number;

[0076] Model integration: All trained independent ANN models are integrated together to form an ANN model cluster with line section numbers.

[0077] Specifically, the disjoint bagging technique is an ensemble learning method that reduces variance and improves the performance of model clustering by dividing the dataset into multiple disjoint subsets and training ANN models separately.

[0078] Model training process: The original training data set is randomly divided into multiple non-overlapping subsets, each of which contains data elements that appear only once in the entire data set, ensuring that the union of these subsets is equal to the original data set.

[0079] An independent artificial neural network (ANN) model is trained on each subset; since the data of each subset is different, the fault characteristics of each line section number learned by each model are also different, thereby improving the diversity of the model.

[0080] Result aggregation: When applying model classification, the output results of all ANN models are integrated, such as taking the majority by voting mechanism, as the final detection result. This aggregation operation can effectively average out the errors and deviations of individual models, thereby reducing the variance of the overall model and improving the stability and accuracy of the model.

[0081] Furthermore, the specific process of using each subset to train an independent ANN model includes:

[0082] The network structure of each ANN model includes an input layer, a hidden layer, and an output layer, and the number and type of neurons in each structural layer are determined; the fault feature vector data set consists of two dimensions, time window and measurement, which are used as neurons in the input layer respectively; the number of neurons in the hidden layer is determined by the Bayesian optimization method; the neurons in the output layer are different fault types;

[0083] Use Bayesian optimization (BO) method to adjust ANN model hyperparameters (such as learning rate, number of iterations, and number of hidden layer neurons); select appropriate activation function and loss function, and use back propagation algorithm to update ANN model parameters;

[0084] Use different subsets to cross-test the performance of each independent ANN model, adjust and optimize the model parameters; until all ANN model performance indicators reach the set convergence value.

[0085] Furthermore, the fault feature vector data set also includes a voltage change rate, that is, a voltage change rate in the same time window; at the same time, the neurons in the input layer also include a voltage change rate.

[0086] Furthermore, the activation function is Sigmoid, Tanh or ReLU function; the loss function is mean square error (MSE) or cross entropy loss function.

[0087] Furthermore, the specific process of adjusting the hyperparameters of the ANN model using the Bayesian optimization method includes:

[0088] a) Use the objective function to define the range of hyperparameters that need to be optimized (such as learning rate, number of hidden layers, number of hidden layer neurons, activation function, etc.); and use the Gaussian process method to model the objective function;

[0089] b) Initialization: randomly select a set of initial hyperparameter combinations and train the ANN model to obtain the corresponding performance indicators;

[0090] c) train and test the corresponding ANN model on the disjoint subsets, and evaluate the performance indicators of the model for the hyperparameter combination;

[0091] d) Estimate the new objective function range: Use the probabilistic surrogate model to estimate the new objective function range based on the existing hyperparameter values ​​and their corresponding performance indicators;

[0092] e) selecting the next hyperparameter combination using an acquisition function according to the new objective function range; the acquisition function includes an expected improvement function or a probability improvement function;

[0093] f) Repeat steps ce) until the model performance index reaches a convergence value.

[0094] Furthermore, the ANN model performance indicators include accuracy, recall and F1 score; the set convergence values ​​are: accuracy ≥ 97%; recall ≥ 95%.

[0095] The fault detection module is used to deploy the trained ANN model cluster; and input the fault feature vector data set extracted by the signal decomposition module into each independent ANN model for fault detection and output the fault type;

[0096] Count the fault types output by each independent ANN model, and select the fault type with the most votes as the final detection result;

[0097] At the same time, among the independent ANN models that meet the final detection results, the line section number with the most votes is selected as the fault location.

[0098] Furthermore, the fault types include short circuit fault, overload fault and ground fault.

[0099] Specifically, the common fault types of distribution network lines mainly include the following:

[0100] Short circuit fault: A low impedance connection occurs between two or more conductors in a circuit, resulting in excessive current between phases.

[0101] Overload failure: The load in the circuit exceeds the rated capacity of the equipment, causing the equipment to overheat or burn.

[0102] Ground Fault: A ground fault occurs in a circuit, causing current to flow to the earth through the ground path.

[0103] The distribution network lines can be divided into sections according to different factors, such as:

[0104] Geographic location: Based on the geographical location, the distribution network lines are divided into different areas, such as urban, suburban, rural, etc.

[0105] Load type: According to the load type, the distribution network lines are divided into different areas, such as residential areas, commercial areas, industrial areas, etc.

[0106] Voltage level: According to the voltage level, the distribution network lines are divided into different areas, such as high voltage, low voltage, etc.

[0107] Fault frequency: Areas with higher fault frequencies should be divided into smaller line sections to facilitate timely isolation of faults.

[0108] Mixed division: Combine multiple division methods and make comprehensive division based on actual conditions.

[0109] In summary, the present invention uses disjoint subsets to train independent ANN models, and then aggregates the classification results of the ANN cluster; thereby reducing errors, alleviating the deviation of a single model, and improving the accuracy and reliability of the overall model cluster.

[0110] The present invention also provides a distribution cabinet for detecting faults and cutting off faulty lines based on big data, which also includes a fault cutting module. The fault cutting module includes a circuit breaker and a main controller. Several circuit breakers are respectively arranged at the end of each line section for cutting off the faulty line; the main controller is respectively communicated with the several circuit breakers for controlling the cutting instructions of the circuit breakers of the corresponding line section according to the detection results of the fault detection module and the fault location.

[0111] Specifically, the circuit breaker is installed at the end of each line section to ensure that the rated current and voltage of the circuit breaker meet the line requirements. For example, Siemens SF6 circuit breakers or GESF6 circuit breakers are selected, which are suitable for high-voltage distribution network lines, especially for high voltage levels and high current capacity occasions.

[0112] Select a suitable master controller and configure the communication protocol to ensure normal communication with the circuit breaker and fault detection module. For example, the ABB PLC series controller can handle complex logic control and data processing. Through automated disconnection operations, the faulty line can be quickly disconnected after a fault occurs, preventing the fault from expanding and ensuring safe and stable operation of the power system.

[0113] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the working process of the modules described above can refer to the corresponding process in the aforementioned specific implementation, and will not be repeated here.

[0114] In the several embodiments provided in the present application, it should be understood that the disclosed power distribution cabinets, modules and processes can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0115] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0116] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

[0117] In the description of the specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0118] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data, characterized in that: It includes signal acquisition module, signal decomposition module, model training module and fault detection module; The signal acquisition module is used to collect the current timing signals of each section line in real time and mark the line section number where the current timing signal occurs; The signal decomposition module is used to perform signal decomposition processing on the collected current time series signal using discrete wavelet transform method to extract the fault feature vector; The model training module is used to build an ANN model cluster; the specific construction process of the model training module includes: Data preparation: Collect current data of various known fault types with different line section numbers and mark the line section numbers; known fault types include short circuit faults, overload faults, and ground faults; Data processing: Use discrete wavelet transform method to perform multi-layer decomposition of the signal, extract the high-frequency components of each layer and calculate the bi-norm; combine the bi-norm groups of each layer into a fault feature vector data set; Dataset division: The fault feature vector dataset is randomly divided into multiple disjoint subsets using disjoint self-aggregation technology, each of which corresponds to a line section number; Model training: Each subset is used to train an independent ANN model, that is, each independent ANN model is labeled with the corresponding line section number; Model integration: All trained independent ANN models are integrated together to form an ANN model cluster with line section numbers; The fault detection module is used to deploy the trained ANN model cluster; and input the fault feature vector data set extracted by the signal decomposition module into each independent ANN model for fault detection and output the fault type; Count the fault types output by each independent ANN model, and select the fault type with the most votes as the final detection result; At the same time, among the independent ANN models that meet the final detection results, the line section number with the most votes is selected as the fault location.

2. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 1, characterized in that: The discrete wavelet transform method is used to perform signal decomposition processing, and the specific process includes: Splitting the current time series signal into current data of multiple fixed time windows; The current data is decomposed into low-frequency components and high-frequency components by using the discrete wavelet transform method and selecting the db-4 wavelet as the wavelet basis function. Extract the high-frequency components at each level and calculate the bi-norm of the high-frequency components. The calculation formula can be expressed as: ； In the formula, is the second norm of the high-frequency component; is the i-th element of the high-frequency component; n is the total number of high-frequency components; The two-norm arrays at each level are synthesized into a fault feature vector dataset , which is used for subsequent signal analysis; its expression is: ； In the formula, is the bi-norm of the i-th element of the high-frequency component; n is the total number of high-frequency components.

3. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 1, characterized in that: The specific process of using each subset to train an independent ANN model includes: The network structure of each ANN model includes an input layer, a hidden layer, and an output layer, and the number and type of neurons in each structural layer are determined; the fault feature vector data set consists of two dimensions, time window and measurement, which are used as neurons in the input layer respectively; the number of neurons in the hidden layer is determined by the Bayesian optimization method; the neurons in the output layer are different fault types; Use Bayesian optimization method to adjust ANN model hyperparameters; select appropriate activation function and loss function, and use back propagation algorithm to update ANN model parameters; Use different subsets to cross-test the performance of each independent ANN model, adjust and optimize the model parameters; until all ANN model performance indicators reach the set convergence value.

4. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 1, characterized in that: The fault feature vector data set in the model training module also includes the voltage change rate, that is, the change rate of the voltage in the same time window; at the same time, the neurons in the input layer of the ANN model also include the voltage change rate.

5. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 3, characterized in that: The activation function is selected from Sigmoid, Tanh or ReLU function; the loss function is selected from mean square error or cross entropy loss function.

6. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 3, characterized in that: The specific process of using the Bayesian optimization method to adjust the ANN model hyperparameters includes: a) Use the objective function to define the range of hyperparameters that need to be optimized; and use the Gaussian process method to model the objective function; b) Initialization: randomly select a set of initial hyperparameter combinations and train the ANN model to obtain the corresponding ANN model performance indicators; c) train and test the corresponding ANN model on the disjoint subsets, and evaluate the performance indicators of the model for the hyperparameter combination; d) Estimate the new objective function range: Use the probabilistic surrogate model to estimate the new objective function range based on the existing hyperparameter values ​​and their corresponding performance indicators; e) selecting the next hyperparameter combination using an acquisition function according to the new objective function range; the acquisition function includes an expected improvement function or a probability improvement function; f) Repeat steps ce) until the model performance index reaches a convergence value.

7. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 6, characterized in that: The performance indicators of the ANN model include accuracy, recall and F1 score; the set convergence values ​​are: accuracy ≥ 97%; recall ≥ 95%.

8. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 1, characterized in that: The fault types include short circuit fault, overload fault and ground fault.

9. A power distribution cabinet for detecting faults and cutting off faulty lines based on big data according to claim 1, characterized in that: It also includes a fault cut-off module, which includes a circuit breaker and a master controller. Several circuit breakers are respectively arranged at the end of each line section to cut off the faulty line; The main controller is connected to the plurality of circuit breakers for communication respectively, and is used to control the cutting instruction of the circuit breaker of the corresponding line section according to the detection result of the fault detection module and the fault position.

Images