A bidirectional power supply intelligent fault diagnosis method, electronic device and storage medium

Abstract

This invention relates to the field of power electronics monitoring technology, and discloses a method, electronic device, and storage medium for intelligent fault diagnosis of bidirectional power supplies. The method includes: acquiring sample signals from the bidirectional power supply under historical operating conditions; performing hierarchical decomposition on the sample signals to obtain multiple sub-band signals of different frequency bands; extracting features from each sub-band signal based on multi-level entropy to obtain complexity features; fusing all complexity features to obtain integrated features; and constructing a fault diagnosis model based on the integrated features and a preset machine learning algorithm to perform online fault diagnosis of the real-time operating conditions of the bidirectional power supply. In this invention, the integration of multi-level entropy combined with a hierarchical decomposition strategy mines multi-band fault information from sample signals from different angles, ensuring the richness and diversity of the acquired fault information, which is beneficial for further improving the accuracy of bidirectional power supply fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power electronic monitoring technology, and in particular to a bidirectional power supply intelligent fault diagnosis method, electronic device, and computer-readable storage medium. Background Technology

[0002] With the continuous development of the new energy industry, bidirectional power supplies account for an increasingly larger proportion of the power system. Considering that any fault in a bidirectional power supply may lead to power outages or voltage fluctuations, thereby affecting the normal operation of user-end equipment, it is necessary to perform real-time fault diagnosis and identification to improve overall reliability and safety. Currently, fault diagnosis methods based on time-domain analysis are mainly used to identify faults in bidirectional power supplies. However, fault signals of complex equipment may be distributed at different angles, and a single time-domain analysis method may not be able to fully uncover the fault characteristics, resulting in a decrease in the accuracy of fault diagnosis. Summary of the Invention

[0003] This invention aims to at least partially address one of the technical problems in related technologies. To this end, this invention proposes a bidirectional power supply intelligent fault diagnosis method, electronic device, and storage medium, which can improve the accuracy of bidirectional power supply fault diagnosis.

[0004] In a first aspect, embodiments of the present invention provide a bidirectional power supply intelligent fault diagnosis method, comprising:

[0005] After obtaining sample signals of the bidirectional power supply under historical operating conditions, the sample signals are hierarchically decomposed to obtain multiple sub-band signals of different frequency bands.

[0006] Feature extraction is performed on each of the sub-band signals based on multi-level entropy to obtain complexity features, and data fusion is performed on all the complexity features to obtain integrated features;

[0007] Based on the integrated features and a preset machine learning algorithm, a fault diagnosis model is constructed to perform online fault diagnosis on the real-time operating conditions of the bidirectional power supply.

[0008] Optionally, in one embodiment of the present invention, the step of hierarchically decomposing the sample signal to obtain multiple sub-band signals of different frequency bands includes:

[0009] Based on the sample signal, a sample sequence corresponding to the sample signal is determined, and a low-frequency operator and a high-frequency operator are constructed according to the sample sequence;

[0010] Construct a layered vector of corresponding dimension based on a predetermined number of layers, wherein the layered node indicated by the layered vector is associated with the number of layers;

[0011] Based on the low-frequency operator, the high-frequency operator, the layer vector, and the layer node, the layer components corresponding to each layer are determined, thereby obtaining multiple sub-band signals of different frequency bands.

[0012] Optionally, in one embodiment of the present invention, the step of extracting features from each of the sub-band signals based on multi-level entropy to obtain complexity features includes:

[0013] Construct a first sub-learner for fuzzy entropy calculation, a second sub-learner for scatter entropy calculation, and a third sub-learner for diversity entropy calculation.

[0014] Based on the first sub-learner, the second sub-learner, and the third sub-learner, entropy values ​​are calculated for the hierarchical components corresponding to each layer to obtain complexity features.

[0015] Optionally, in one embodiment of the present invention, the step of calculating entropy values ​​for the hierarchical components corresponding to each layer based on the first sub-learner, the second sub-learner, and the third sub-learner to obtain complexity features includes:

[0016] Multiple sub-feature signal groups are pre-constructed, wherein the multiple sub-feature signal groups include a first sub-feature signal group, a second sub-feature signal group, and a third sub-feature signal group;

[0017] According to the number of layers of each layered component, each layered component is divided into one of the corresponding sub-feature signal groups;

[0018] Each of the hierarchical components in the first sub-feature signal group is input into the first sub-learner to obtain multiple first entropy values; each of the hierarchical components in the second sub-feature signal group is input into the second sub-learner to obtain multiple second entropy values; and each of the hierarchical components in the third sub-feature signal group is input into the third sub-learner to obtain multiple third entropy values.

[0019] All the first entropy values, all the second entropy values, and all the third entropy values ​​are used as complexity features.

[0020] Optionally, in one embodiment of the present invention, the step of constructing a fault diagnosis model based on the integrated features combined with a preset machine learning algorithm includes:

[0021] Based on the integrated features, the initial extreme learning machine is subjected to loss minimization learning to obtain the intermediate state of the extreme learning machine;

[0022] The integrated features are input into the intermediate state of the extreme learning machine for training to obtain the pre-trained extreme learning machine;

[0023] The pre-trained extreme learning machine is used as the corresponding fault diagnosis model.

[0024] Optionally, in one embodiment of the present invention, the step of inputting each of the hierarchical components in the second sub-feature signal group into the second sub-learner to obtain a plurality of second entropy values ​​includes:

[0025] For each of the hierarchical components input into the second sub-learner, the hierarchical component is mapped to a first signal sequence through a normal probability cumulative distribution function;

[0026] For each feature element in the first signal sequence, a linear classification algorithm is used to assign one of the pre-configured different dispersion class numbers to the feature element, and based on the assigned dispersion class number and the pre-set embedding dimension and delay factor, a feature embedding vector corresponding to the feature element is constructed.

[0027] Calculate all dispersion modes corresponding to the feature embedding vector, and calculate the probability of each dispersion mode respectively;

[0028] The second entropy value of the layered component is obtained by combining the information entropy with the probabilities of various dispersion modes.

[0029] Optionally, in one embodiment of the present invention, the step of inputting each of the hierarchical components in the third sub-feature signal group into the third sub-learner to obtain multiple third entropy values ​​includes:

[0030] For each of the hierarchical components input into the third sub-learner, phase space reconstruction is performed based on the hierarchical components to obtain multiple orbital sequences;

[0031] Calculate the cosine similarity between any two adjacent orbital sequences, and generate a similarity vector based on all the cosine similarities;

[0032] The target interval is divided into multiple sub-intervals, and the histcounts function is used to count the similarity vector in each sub-interval according to the multiple sub-intervals as intervals, so as to obtain the statistical probability of the similarity vector in each sub-interval;

[0033] The third entropy value of the hierarchical component is obtained by combining the information entropy with the statistical probability of the similarity vector in each sub-interval.

[0034] In a second aspect, embodiments of the present invention provide an electronic device, comprising:

[0035] At least one processor;

[0036] At least one memory for storing at least one program;

[0037] The bidirectional power supply intelligent fault diagnosis method as described in the first aspect is implemented when at least one of the programs is executed by at least one of the processors.

[0038] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the bidirectional power supply intelligent fault diagnosis method as described in the first aspect.

[0039] This invention proposes a bidirectional power supply intelligent fault diagnosis method, electronic device, and storage medium. It decomposes sample signals into sub-band signals of different frequency bands using a hierarchical decomposition strategy, and then extracts features from each sub-band signal based on multi-level entropy. This ensures the richness and diversity of fault information obtained. Furthermore, it fuses all the complex features to obtain integrated features with higher accuracy that characterize the bidirectional power supply's operating conditions, thus improving the accuracy of bidirectional power supply fault diagnosis. Based on this, a fault diagnosis model is constructed using the integrated features and a pre-set machine learning algorithm. This model enables online fault diagnosis of the real-time operating conditions of the bidirectional power supply, further improving the accuracy of bidirectional power supply fault diagnosis. Attached Figure Description

[0040] Figure 1 This is a flowchart of a bidirectional power supply intelligent fault diagnosis method provided in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the execution steps of a bidirectional power supply intelligent fault diagnosis method according to an embodiment of the present invention;

[0042] Figure 3 yes Figure 1 The flowchart of step S1 in the process;

[0043] Figure 4 yes Figure 1 The flowchart of step S2 in the process of “extracting features from each sub-band signal based on multi-level entropy to obtain complexity features” is a partial flowchart.

[0044] Figure 5 yes Figure 4 The flowchart of step S22 in the text;

[0045] Figure 6 yes Figure 4 A schematic diagram of the execution flow of step S22 in the diagram;

[0046] Figure 7 yes Figure 5The flowchart of step S223, "Input each hierarchical component in the second sub-feature signal group into the second sub-learner to obtain multiple second entropy values ​​respectively" is shown in part.

[0047] Figure 8 yes Figure 5 The flowchart of step S223, "Input each hierarchical component in the third sub-feature signal group into the third sub-learner to obtain multiple third entropy values ​​respectively" is shown in part.

[0048] Figure 9 yes Figure 8 A diagram illustrating the execution steps;

[0049] Figure 10 This is a schematic diagram of the execution steps of a bidirectional power supply intelligent fault diagnosis method provided in another embodiment of the present invention;

[0050] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0051] Figure 1 This is a flowchart illustrating a bidirectional power supply intelligent fault diagnosis method according to an embodiment of the present invention. Figure 1 As shown, the bidirectional power supply intelligent fault diagnosis method may include, but is not limited to, steps S1 to S3.

[0052] Step S1: After acquiring sample signals of the bidirectional power supply under historical operating conditions, the sample signals are hierarchically decomposed to obtain multiple sub-band signals of different frequency bands. The historical operating conditions characterize the working state of the bidirectional power supply under historical operating conditions; therefore, the collected sample signals are the historical operating signals of the bidirectional power supply. The specific values ​​and types of these sample signals can be set according to the specific scenario, for example, referring to... Figure 2 It can sample voltage, current, frequency and temperature signals under different historical operating conditions, that is, for the capacity testing scenario, the bidirectional power supply's operating data under the capacity testing scenario can be stored in a pre-set scenario database in real time, so that when a sampling signal is needed, it can be directly obtained from the scenario database.

[0053] Step S2: Extract features from each sub-band signal based on multi-level entropy to obtain complexity features, and fuse all complexity features to obtain integrated features;

[0054] Step S3: Construct a fault diagnosis model based on integrated features and a preset machine learning algorithm, so as to perform online fault diagnosis on the real-time operating conditions of the bidirectional power supply through the fault diagnosis model.

[0055] In this step, the sample signal is decomposed into sub-band signals of different frequency bands through a hierarchical decomposition strategy. Then, feature extraction is performed on each sub-band signal based on multi-level entropy, which ensures the richness and diversity of fault information. Data fusion is performed on all complex features to obtain integrated features with higher accuracy that can characterize the bidirectional power supply operating conditions. This is beneficial to improving the accuracy of bidirectional power supply fault diagnosis. On this basis, a fault diagnosis model is constructed based on the integrated features and a preset machine learning algorithm. This allows for online fault diagnosis of the real-time operating conditions of the bidirectional power supply, further improving the accuracy of bidirectional power supply fault diagnosis.

[0056] In one embodiment, before performing hierarchical decomposition on the sample signal, the sample signal may be subjected to noise reduction processing, but not limited to noise removal, in order to remove redundant features in the sample signal, reduce the impact of noise on feature extraction, and thus improve the accuracy of feature extraction.

[0057] like Figure 3 As shown, in one embodiment of the present invention, step S1 may include, but is not limited to, the following steps:

[0058] Step S11: Determine the sample sequence corresponding to the sample signal based on the sample signal, and construct low-frequency operators and high-frequency operators based on the sample sequence;

[0059] Step S12: Construct a layered vector of the corresponding dimension according to the predetermined number of layers, wherein the layered nodes indicated by the layered vector are associated with the number of layers;

[0060] Step S13: Based on the low-frequency operator, high-frequency operator, layer vector, and layer node, determine the corresponding layer components for each layer, thereby obtaining multiple sub-band signals of different frequency bands.

[0061] In this step, considering complex fault conditions, fault information may exist not only in the low-frequency components of the signal but also in the high-frequency components in a multi-scale manner. However, existing time-domain analysis methods mainly utilize dimensionless parameters for feature extraction, failing to uncover the simultaneous presence of both high-frequency and low-frequency components. Therefore, as... Figure 2 As shown, a hierarchical decomposition strategy is used to divide the collected sample sequence. First, low-frequency and high-frequency operators are constructed based on the sample sequence. Then, a hierarchical vector of the corresponding dimension is constructed according to the predetermined number of hierarchical layers. Based on this, the hierarchical components corresponding to each layer can be determined by combining the low-frequency operators, high-frequency operators, hierarchical vectors, and hierarchical nodes. Each layer corresponds to a different frequency band, thus obtaining multiple sub-band signals of different frequency bands.

[0062] Specifically, the principle of hierarchical decomposition can be summarized, but is not limited to, as follows:

[0063] Assume the sample sequence is of length 2 n sequence Low-frequency operators can then be constructed separately. and high-frequency operators It can be, but is not limited to, for

[0064] ;

[0065] Or, operators , It can also be represented by the following matrix:

[0066] ;

[0067] Then, select the number of layers. , Indicates the maximum number of layers that can be layered, and constructs... dimensional vector ,in, 0 or , That is, the hierarchical nodes can be calculated using the following formula:

[0068] ;

[0069] for dimensional vector Repeat the above steps to obtain the first one. Layered nodes The corresponding layered component, that is, the first The sub-band signals corresponding to the frequency band of the layer are shown below:

[0070] .

[0071] like Figure 4 As shown, in one embodiment of the present invention, step S2, which involves extracting features from each sub-band signal based on multi-level entropy to obtain complexity features, may include, but is not limited to, the following steps:

[0072] Step S21: Construct a first sub-learner for performing fuzzy entropy calculation, a second sub-learner for performing scatter entropy calculation, and a third sub-learner for performing diversity entropy calculation, respectively.

[0073] Step S22: Based on the first sub-learner, the second sub-learner, and the third sub-learner, calculate the entropy value of the corresponding hierarchical components of each layer to obtain the complexity features.

[0074] In this step, the complexity of sub-band signals is evaluated from the perspective of entropy. By constructing multiple sub-learners to extract the entropy features of different sub-band signals, data fusion of entropy features of different frequency bands can be performed. Compared with signal analysis and processing methods in related technologies, the entropy algorithm has lower requirements for signals. For nonlinear and non-stationary vibration signals, the entropy algorithm can better extract the feature information of the signal. When a fault occurs, the internal complexity of the signal will change. At this time, the change in the entropy value of the signal can indicate the change in the system state. Therefore, based on the first, second, and third sub-learners, the entropy value is calculated for the corresponding layered components of each layer, and the complexity features under bidirectional power supply conditions can be extracted from multiple perspectives.

[0075] like Figure 5 As shown, in one embodiment of the present invention, step S22 may include, but is not limited to, the following steps:

[0076] Step S221: Preconstruct multiple sub-feature signal groups, wherein the multiple sub-feature signal groups include a first sub-feature signal group, a second sub-feature signal group, and a third sub-feature signal group;

[0077] Step S222: According to the number of layers of each layer component, divide each layer component into one of the corresponding sub-feature signal groups;

[0078] Step S223: Input each layered component in the first sub-feature signal group into the first sub-learner to obtain multiple first entropy values; input each layered component in the second sub-feature signal group into the second sub-learner to obtain multiple second entropy values; and input each layered component in the third sub-feature signal group into the third sub-learner to obtain multiple third entropy values.

[0079] Step S224: Use all first entropy values, all second entropy values, and all third entropy values ​​as complexity features.

[0080] In this step, fuzzy entropy measures signal complexity by the probability of generating new patterns within the signal; dispersion entropy (DE), also known as dispersive entropy, uses the probability of reconstructing the internal arrangement order of subsequences as the basis for entropy value calculation; and diversity entropy (DivE) calculates information entropy by the statistical probability of cosine similarity between different tracks of the signal. The change in angle within the signal measures signal complexity. All three are characterized by fast calculation and good stability. Therefore, as... Figure 2As shown, based on the hierarchical decomposition strategy, feature integration is performed using fuzzy entropy, scatter entropy, and diversity entropy as three sub-learners to obtain the first, second, and third sub-learners respectively. Each hierarchical component is adaptively grouped so that each hierarchical component can be input into the corresponding sub-learner for entropy calculation, resulting in the corresponding entropy value. The larger the entropy value, the higher the signal complexity. Overall, the health of the bidirectional power source can be measured from three aspects: the probability of signal mutation, the internal arrangement order of the signal, and angular information. That is, all first entropy values, all second entropy values, and all third entropy values ​​are used as the complexity features of the bidirectional power source.

[0081] In one embodiment, each layer component is divided into one of its corresponding sub-feature signal groups according to the number of layers of each layer component. This can be implemented according to the actual application scenario, that is, depending on the division of the first sub-feature signal group, the second sub-feature signal group, and the third sub-feature signal group. For example, one or more layer thresholds can be set to divide multiple threshold intervals, and each threshold interval corresponds to one of the sub-feature signal groups. The number of layers of each layer component is traversed, and the number of layers of each layer component is divided into one of its corresponding sub-feature signal groups according to whether the number of layers is within the threshold interval.

[0082] Specifically, such as Figure 6 As shown, with the hierarchical thresholds set to 1, 2, and 3, when calculating the hierarchical entropy, when... When the sample signal is divided into high-frequency and low-frequency sub-components, the calculated entropy value represents the initial state and overall fluctuation of the initial signal. In this case, the first sub-learner is used to calculate the entropy value; when At this point, the sample signal is decomposed into two high-frequency components and two low-frequency components. The calculated entropy value at this time represents the detailed information of the initial signal. The second sub-learner is selected to extract features from four components. Similarly, with Taking a value equal to 3 as an example, the sample signal is decomposed into four high-frequency components and four low-frequency components. A third sub-learner is selected to calculate the entropy value. Finally, the complexity features extracted in the above embodiments are fused using the Conact data fusion method. That is, by balancing fault information from different perspectives, the accuracy of judging the fault problem results is enhanced. The corresponding extracted feature length is 2. The corresponding extracted feature length is 4. The feature length extracted at that time was 8, so the feature length finally extracted based on the Conact data fusion method was 14.

[0083] like Figure 7As shown, in one embodiment of the present invention, step S223, which involves inputting each hierarchical component of the second sub-feature signal group into the second sub-learner to obtain multiple second entropy values, may include, but is not limited to, the following steps:

[0084] Step S2231: For each hierarchical component input into the second sub-learner, the hierarchical component is mapped to the first signal sequence through the normal distribution probability cumulative distribution function;

[0085] Step S2232: For each feature element in the first signal sequence, a linear classification algorithm is used to assign one of the corresponding dispersion categories to the feature element from a pre-configured number of different dispersion categories. Based on the assigned number of dispersion categories and the pre-set embedding dimension and delay factor, a feature embedding vector corresponding to the feature element is constructed.

[0086] Step S2233: Calculate all dispersion modes corresponding to the feature embedding vector, and calculate the probability of each dispersion mode respectively;

[0087] Step S2234: Based on the information entropy combined with the probabilities of various dispersion modes, the second entropy value of the layered component is obtained.

[0088] Specifically, for a univariate signal of length N The calculation process for its scatter entropy is as follows:

[0089] First, Mapped to signal ,Right now

[0090] ;

[0091] in, express The first in One operation, express variance express The mean;

[0092] Furthermore, select the largest number of dispersion categories. Embedded Dimension and delay factor For each mapped element Using linear classification algorithms in related technologies (which can be implemented using a linear classifier, training a model with features and labels, and finding a linear dividing line to separate data points into two classes) from 1 to... The middle element is the element. Assign a corresponding level, and then construct the embedding vector. ;

[0093] Furthermore, calculate the dispersion mode. ,make , … Then we can know that The corresponding dispersion mode is ,because Related to Number of digits, and each number exists The possible values ​​are: therefore, the corresponding dispersion modes are: indivual;

[0094] Further calculation probability Specifically:

[0095] ;

[0096] in, express Mapped to The number of That is, characterization Mapped to The number of and The quotient of the number of elements in the matrix;

[0097] Finally, based on the information entropy and the probabilities of the dispersive modes calculated above, the dispersive entropy is calculated. As shown below:

[0098] .

[0099] like Figure 8 As shown, in one embodiment of the present invention, step S223, which involves inputting each hierarchical component of the third sub-feature signal group into the third sub-learner to obtain multiple third entropy values, may include, but is not limited to, the following steps:

[0100] Step S2235: For each hierarchical component input into the third sub-learner, the phase space is reconstructed based on the hierarchical component to obtain multiple orbital sequences;

[0101] Step S2236: Calculate the cosine similarity between any two adjacent orbital sequences, and generate a similarity vector based on all cosine similarities;

[0102] Step S2237: Divide the target interval into multiple sub-intervals, and use the histcounts function to count the similarity vector in each sub-interval according to the interval of multiple sub-intervals, so as to obtain the statistical probability of the similarity vector in each sub-interval;

[0103] Step S2238: Based on the information entropy and the statistical probability of the similarity vector in each sub-interval, obtain the third entropy value of the hierarchical component.

[0104] Specifically, such as Figure 9 As shown, for a sequence of length N The calculation process for its diversity entropy is as follows:

[0105] First, the sequence undergoes phase space reconstruction. The purpose of this reconstruction is to extract more information from the entire sequence and find another dynamical system equivalent to the original sequence. The dynamical characteristics are then studied by reconstructing the phase space orbits. Based on phase space embedding theory, and given the embedding dimension and delay factor, the sequence... It can be reconstructed into a series of orbits, which can be represented as ;

[0106] Then, calculate the adjacent orbital sequences. and cosine similarity Then, all the calculation results are combined to obtain a similarity vector. Specifically, cosine similarity for:

[0107] ;

[0108] Furthermore, the target interval corresponding to the similarity is selected as [-1, 1], and this interval is divided into... There are subintervals, and each subinterval can be represented as... Therefore, based on the divided sub-intervals, the histcounts function (well-known to those skilled in the art and will not be elaborated here) is used to count the similarity vectors in each sub-interval, and the statistical probability of the similarity vectors in each sub-interval is calculated. ;

[0109] Finally, based on the definition of information entropy and the statistical probability of the similarity vector within each sub-interval, the entropy is calculated and normalized to obtain the diversity entropy:

[0110] ;

[0111] As can be seen from the above calculation process, the core idea of ​​diversity entropy lies in measuring signal complexity and the number of intervals by calculating the statistical probability of the cosine similarity between the internal orbits of the signal. This determines the level of detail in the entropy calculation process.

[0112] In one embodiment of the present invention, step S3 may include, but is not limited to, the following steps:

[0113] Step S31: Minimize the loss learning of the initial extreme learning machine based on the ensemble features to obtain the intermediate state extreme learning machine;

[0114] Step S32: Input the integrated features into the intermediate state of the extreme learning machine for training to obtain a pre-trained extreme learning machine;

[0115] Step S33: Use the pre-trained extreme learning machine as the corresponding fault diagnosis model.

[0116] Considering the high requirements for fault response speed, compared with traditional machine learning algorithms, Extreme Learning Mainshine (ELM) avoids iterative adjustment of network weights, saving time costs and effectively solving the problems of high dimensionality, big data, and training difficulty. ELM is a single-hidden-layer feedforward neural network. First, it randomly generates the input weights and biases of hidden layer neurons, and then calculates the output matrix of the single-hidden-layer feedforward neural network through a regularized linear regression method. Even though the hidden layer weights are randomly generated, ELM still maintains the general approximation ability of a single-hidden-layer feedforward neural network. By randomly generating the parameters of the hidden layer nodes, it achieves fast training speed, good efficiency in processing large-scale data, and fast learning speed. It also has excellent performance in handling nonlinear relationships and complex problems. Its principle is well known to those skilled in the art and will not be elaborated here. Please refer to https: / / cloud.tencent.com / developer / article / 2113705 and https: / / blog.csdn.net / weixin_32747681 / article / details / 149856503 for more information.

[0117] In this step, such as Figure 2 , Figure 10As shown, in offline mode, the initial Extreme Learning Machine (ELM) can be trained using the ensemble features in the feature pool as the training set to minimize the loss, resulting in an intermediate-state ELM model. The ensemble features are then input into this intermediate-state ELM for training, yielding the trained parameters. Based on this, the pre-trained ELM can be determined, which is the corresponding fault diagnosis model. After determining the fault diagnosis model, the process returns to online mode. Real-time data is sampled during bidirectional power supply operation, and the collected data undergoes hierarchical ensemble entropy feature extraction (the principle is the same as in the above embodiments, the only difference being that the base features are used in this case). Feature extraction is performed on real-time sampled data (which will not be elaborated here), and then the data is used as a test set to be input into the Extreme Learning Machine (ELM). The ELM is used to classify faults and output the corresponding fault type. If the output is an unknown fault type, the classification samples and fault detection results can be saved back to the scene database corresponding to the historical data of the breakdown and capacity, which is used as the new historical data of the breakdown and capacity. This allows for the updating of the ELM parameters offline based on the updated historical data of the breakdown and capacity. It can be understood that even if the output is a known fault type, the corresponding classification data can still be saved back to the scene database corresponding to the historical data of the breakdown and capacity. There is no restriction on this.

[0118] Figure 11 This is a schematic diagram of the structure of an electronic device 1000 provided in an embodiment of the present invention. For example... Figure 11 As shown, the electronic device 1000 includes a memory 1100 and a processor 1200. The number of memories 1100 and processors 1200 can be one or more. Figure 11 Taking a memory 1100 and a processor 1200 as an example; the memory 1100 and the processor 1200 in the device can be connected via a bus or other means. Figure 11 Taking the example of a connection between China and Israel via a bus.

[0119] The memory 1100, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the bidirectional power supply intelligent fault diagnosis method provided in any embodiment of the present invention. The processor 1200 implements the above-described bidirectional power supply intelligent fault diagnosis method by running the software programs, instructions, and modules stored in the memory 1100.

[0120] The memory 1100 may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system and application programs required for at least one function. Furthermore, the memory 1100 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 1100 may further include memory remotely located relative to the processor 1200, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0121] An embodiment of the present invention also provides a computer-readable storage medium storing computer-executable instructions for performing the bidirectional power supply intelligent fault diagnosis method provided in any embodiment of the present invention.

[0122] An embodiment of the present invention also provides a computer program product, including a computer program or computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer program or computer instructions from the computer-readable storage medium and executes the computer program or computer instructions, causing the computer device to perform the bidirectional power supply intelligent fault diagnosis method provided in any embodiment of the present invention.

[0123] The electronic devices and application scenarios described in the embodiments of this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of electronic devices and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0124] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0125] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor (such as a central processing unit, digital signal processor, or microprocessor), or as hardware, or as integrated circuits, such as application-specific integrated circuits (ASICs). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible by a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0126] The terms “component,” “module,” “system,” etc., used in this specification are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process or execution thread, and components may be located on a single computer or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, or a network, such as the Internet interacting with other systems via signals).

Claims

1. A bidirectional power supply intelligent fault diagnosis method, characterized in that, include: After obtaining sample signals of the bidirectional power supply under historical operating conditions, the sample signals are hierarchically decomposed to obtain multiple sub-band signals of different frequency bands. Feature extraction is performed on each of the sub-band signals based on multi-level entropy to obtain complexity features, and data fusion is performed on all the complexity features to obtain integrated features; Based on the integrated features and a preset machine learning algorithm, a fault diagnosis model is constructed to perform online fault diagnosis on the real-time operating conditions of the bidirectional power supply. The step of performing hierarchical decomposition on the sample signal to obtain multiple sub-band signals of different frequency bands includes: Based on the sample signal, a sample sequence corresponding to the sample signal is determined, and a low-frequency operator and a high-frequency operator are constructed according to the sample sequence; Construct a layered vector of corresponding dimension based on a predetermined number of layers, wherein the layered node indicated by the layered vector is associated with the number of layers; Based on the low-frequency operator, the high-frequency operator, the layer vector, and the layer node, the layer components corresponding to each layer are determined respectively, thereby obtaining multiple sub-band signals of different frequency bands; The step of extracting features from each of the sub-frequency band signals based on multi-level entropy to obtain complexity features includes: Construct a first sub-learner for fuzzy entropy calculation, a second sub-learner for scatter entropy calculation, and a third sub-learner for diversity entropy calculation. Multiple sub-feature signal groups are pre-constructed, wherein the multiple sub-feature signal groups include a first sub-feature signal group, a second sub-feature signal group, and a third sub-feature signal group; According to the number of layers of each layered component, each layered component is divided into one of the corresponding sub-feature signal groups; Each of the hierarchical components in the first sub-feature signal group is input into the first sub-learner to obtain multiple first entropy values; each of the hierarchical components in the second sub-feature signal group is input into the second sub-learner to obtain multiple second entropy values; and each of the hierarchical components in the third sub-feature signal group is input into the third sub-learner to obtain multiple third entropy values. All the first entropy values, all the second entropy values, and all the third entropy values ​​are used as complexity features.

2. The bidirectional power supply intelligent fault diagnosis method according to claim 1, characterized in that, The construction of the fault diagnosis model based on the integrated features and a preset machine learning algorithm includes: Based on the integrated features, the initial extreme learning machine is subjected to loss minimization learning to obtain the intermediate state of the extreme learning machine; The integrated features are input into the intermediate state of the extreme learning machine for training to obtain the pre-trained extreme learning machine; The pre-trained extreme learning machine is used as the corresponding fault diagnosis model.

3. The bidirectional power supply intelligent fault diagnosis method according to claim 1, characterized in that, The step of inputting each of the hierarchical components in the second sub-feature signal group into the second sub-learner to obtain multiple second entropy values ​​includes: For each of the hierarchical components input into the second sub-learner, the hierarchical component is mapped to a first signal sequence through a normal probability cumulative distribution function; For each feature element in the first signal sequence, a linear classification algorithm is used to assign one of the pre-configured different dispersion class numbers to the feature element, and based on the assigned dispersion class number and the pre-set embedding dimension and delay factor, a feature embedding vector corresponding to the feature element is constructed. Calculate all dispersion modes corresponding to the feature embedding vector, and calculate the probability of each dispersion mode respectively; The second entropy value of the layered component is obtained by combining the information entropy with the probabilities of various dispersion modes.

4. The bidirectional power supply intelligent fault diagnosis method according to claim 1, characterized in that, The step of inputting each of the hierarchical components in the third sub-feature signal group into the third sub-learner to obtain multiple third entropy values ​​includes: For each of the hierarchical components input into the third sub-learner, phase space reconstruction is performed based on the hierarchical components to obtain multiple orbital sequences; Calculate the cosine similarity between any two adjacent orbital sequences, and generate a similarity vector based on all the cosine similarities; The target interval is divided into multiple sub-intervals, and the histcounts function is used to count the similarity vector in each sub-interval according to the multiple sub-intervals as intervals, so as to obtain the statistical probability of the similarity vector in each sub-interval; The third entropy value of the hierarchical component is obtained by combining the information entropy with the statistical probability of the similarity vector in each sub-interval.

5. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; The bidirectional power supply intelligent fault diagnosis method as described in any one of claims 1 to 4 is implemented when at least one of the programs is executed by at least one of the processors.

6. A computer-readable storage medium, characterized in that, It stores a processor-executable program, which, when executed by the processor, is used to implement the bidirectional power supply intelligent fault diagnosis method as described in any one of claims 1 to 4.

Images