Charging pile fault detection and diagnosis method and device based on neural network, and medium

Abstract

The invention relates to a charging pile fault detection and diagnosis method and device based on a neural network and a medium, and belongs to the field of charging fault detection. The method comprises the steps of collecting various fault features of the charging pile; corresponding sub-classifiers are constructed for various fault features, and the sub-classifiers are weak classifiers for fault prediction based on the corresponding fault features; and evidence reasoning is carried out, output results of the sub-classifiers are fused, and a final fault diagnosis result fusing prediction diagnosis results of the multiple sub-classifiers is obtained. Weak classifiers for performing fault analysis by applying fault features of different modes are combined through evidence reasoning, and uncertainty and incomplete information can be well processed through evidence reasoning; probability information can be processed, and uncertainty wider than probability can be processed; information from different sub-classifiers is effectively merged, and a reasonable synthesis result can be given even if conflicts exist between evidences. And the method has certain robustness for noise or errors of a single evidence source.

Description

Technical Field

[0001] The present invention relates to the technical field of charging pile fault diagnosis, and more specifically, to a method, device and medium for charging pile fault detection and diagnosis based on a neural network. Background Art

[0002] With the large-scale development of new energy electric vehicles, the construction of supporting charging infrastructure has also followed closely. The service life and operation reliability of charging equipment directly affect the charging experience of customers. Among them, the charging pile, as the core spare part of the charging pile, is directly related to charging customers. Due to the high frequency of use of the charging pile and the long-term working scenario of high voltage and large current, the proportion of charging pile faults in charging equipment faults is as high as more than 80%.

[0003] Common fault diagnosis methods include: Fault diagnosis method based on signal processing: This method is applied to the situation where it is difficult to establish a mathematical analytical model for the diagnostic object but there are input and output signals. Its essence is to collect signals, establish a signal model, analyze the measurable signals using this model, extract feature values, and perform fault identification. Expert system-based diagnosis method: The expert system is a diagnosis method in the field of artificial intelligence. It does not require the establishment of a mathematical model or signal processing. It is an intelligent computer program designed based on the long-term practical experience of experts and the collected fault information. It sorts out the fault information of the diagnostic object and stores it in the expert system knowledge base. According to the characteristics corresponding to the faults, a certain reasoning method is adopted to find out the cause of the fault. This method can achieve rapid fault diagnosis and has a significant effect on complex diagnostic objects. However, due to the difficulty of knowledge acquisition in the knowledge base, the application of this method is limited. Fault diagnosis method based on neural network: Fault diagnosis based on neural network establishes the connection between data and fault types through a large number of data samples, and gives a diagnostic result based on the established connection. Fusing weak classifiers for fault diagnosis is an important branch of fault diagnosis based on neural network. Common fusions include: Stacking algorithm: The stacking algorithm constructs a multi-layer structure of classifiers, makes better use of the information of different classifiers, and improves the accuracy. However, the multi-layer classifier has a high computational cost and a complex structure; Boosting algorithm: The boosting algorithm iteratively trains weak classifiers to construct a strong classifier. During the process, by focusing on misclassified samples, the sample weights are automatically adjusted, so that the network pays more attention to the wrong samples, but it is greatly affected by noise and the adaboost computational cost is relatively high; Bagging algorithm: The bagging algorithm reduces the model error through random sampling and voting. However, for models with large deviations, the performance improvement brought by bagging is not obvious. Summary of the Invention

[0004] To solve the above technical problems or at least partially solve the above technical problems, the present invention provides a method, device and medium for detecting and diagnosing charging pile faults based on a neural network.

[0005] In a first aspect, the present invention provides a method for detecting and diagnosing charging pile faults based on a neural network, including:

[0006] Collect various fault features of the charging pile; the fault features include: current fault texture features, voltage fault texture features, and temperature features of the set components of the charging pile;

[0007] For various fault features, construct corresponding sub-classifiers, and the sub-classifiers are weak classifiers for fault prediction based on the corresponding fault features;

[0008] Perform evidence reasoning, fuse the output results of each sub-classifier, and obtain the final fault diagnosis result that combines the prediction and diagnosis results of multiple sub-classifiers.

[0009] Furthermore, the current fault texture features adopt any one or a combination of several of the following types:

[0010] Statistical current fault texture features: current average value, standard deviation, smoothness, skewness, kurtosis, and root mean square value;

[0011] Frequency-domain current fault texture features: current frequency-domain features after Fourier transform of current data;

[0012] Time-frequency domain current fault texture features: current time-frequency domain features obtained by wavelet transform of current data;

[0013] The voltage fault texture features adopt one or a combination of several of the following types:

[0014] Statistical voltage fault texture features: voltage average value, standard deviation, smoothness, skewness, kurtosis, and root mean square value;

[0015] Frequency-domain voltage fault texture features: voltage frequency-domain features after Fourier transform of voltage data;

[0016] Time-frequency domain voltage fault texture features: voltage time-frequency domain features obtained by wavelet transform of voltage data.

[0017] Furthermore, constructing corresponding sub-classifiers for various fault features includes: setting the kth type of fault feature X k , and the sub-classifier for the kth type of fault feature X k is C k , and the calculation method of the classifier weights of all sub-classifiers is as follows:

[0018] Generate feature nodes by randomly mapping and sparsely encoding fault features:

[0019]

[0020] Among them, is the m-th feature node corresponding to the sub-classifier C k , is the corresponding random weight matrix, is the corresponding random bias, a 1 () is the non-linear activation function, M is the total number of feature nodes, and K is the total number of sub-classifiers;

[0021] Concatenate all the feature nodes of each sub-classifier:

[0022]

[0023] Perform further random mapping and sparse encoding on the concatenation results of all the feature nodes of each sub-classifier to generate enhanced feature nodes:

[0024]

[0025] Among them, is the n-th enhanced feature node corresponding to the sub-classifier C k , is the corresponding random weight matrix, is the corresponding random bias, a 2 () is the non-linear activation function, and N is the total number of enhanced feature nodes;

[0026] Concatenate all the enhanced feature nodes of each type of sub-classifier

[0027] Concatenate the concatenation results of all the feature nodes and the concatenation results of the enhanced feature nodes of each type of sub-classifier into the input matrix of each type of sub-classifier, and calculate the weights of each type of sub-classifier through ridge regression based on the input matrix and the charging pile fault type labels:

[0028]

[0029] Among them, W k is the weight of the k-th sub-classifier, is the input matrix of each type of sub-classifier pseudo-inverse matrix, and Y is the charging pile fault type label.

[0030] Furthermore, evidence reasoning is carried out to fuse the output results of each sub-classifier to obtain the final fault diagnosis result that fuses the predicted diagnosis results of multiple sub-classifiers, including:

[0031] Determine the recognition range as Y, where Y is the set of charging pile fault type labels, Y = {Y 1 , Y 2 …, Y i , …Y I}, where Y i represents the i-th charging pile fault type label;

[0032] Each sub-classifier assigns confidence levels to various fault characteristics for predicting various fault type labels

[0033] Calculate the global uncertainty based on the confidence levels Global uncertainty represents the uncertain part of the evidence;

[0034] Assign weights to various charging pile fault type predictions of each sub-classifier according to the accuracy of the predictions of each sub-classifier for various charging pile fault type labels

[0035] The results of various charging pile fault type predictions of each sub-classifier are weighted by the weights of various charging pile fault type predictions of each sub-classifier to obtain the basic probability mass of the evidence for each fault category, and calculate the probability mass of the uncertain part;

[0036] Recursively fuse the basic probability mass and the probability mass of the uncertain part of two pieces of evidence;

[0037] Calculate the synthesized confidence distribution Among them, the basic probability mass of each synthesized charging pile fault type is The probability mass of the uncertain part of each synthesized charging pile fault type is β Y ;

[0038] According to the synthesized confidence distribution Select the charging pile fault type as the final fault diagnosis result.

[0039] Furthermore, based on the sub-classifier weights W k , k = 1, 2,..., K of the current iteration step and the input matrix, calculate the predicted labels of all charging pile fault types for each sub-classifier Where I is the total number of charging pile fault type labels; determine the accuracy of each type of charging pile fault type label predicted by each sub-classifier, and allocate the contribution of each type of charging pile fault type predicted by the sub-classifier according to the accuracy of each type of charging pile fault type label predicted by the sub-classifier

[0040] Where acc() is the accuracy function, is the accuracy of the k-th sub-classifier predicting the i-th type of charging pile fault type label.

[0041] Furthermore, the weights of each type of charging pile fault type predicted by each sub-classifier are weighted by the results of each sub-classifier predicting each type of charging pile fault type to obtain the basic probability mass of the evidence for each fault category:

[0042]

[0043] Then the probability mass of the uncertain part is:

[0044]

[0045] The probability mass of the uncertain part consists of two parts: the unallocated probability due to weights and the original uncertain part of the evidence

[0046] Furthermore, the recursive fusion of the basic probability masses of two pieces of evidence includes: recursively combining the basic probability masses of two pieces of evidence into a new basic probability mass using the combination formula, combining the probability masses of the uncertain parts of two pieces of evidence into a new probability mass of the uncertain part, and repeating this process until the basic probability masses and the probability masses of the uncertain parts of all pieces of evidence are combined. The combination formula includes:

[0047] Combining two basic probability masses:

[0048]

[0049] Calculating the probability mass of the combined uncertain part:

[0050]

[0051] Finally,

[0052] where λ is the normalization factor, subtracting the sum of the products of the basic probability masses of all conflict terms. All conflict terms, that is, all combinations of different charging pile fault types ij, and the basic probability mass of each charging pile fault type after combination is The probability mass of the uncertain part of each charging pile fault type after synthesis is β Y 。

[0053] Furthermore, for the current fault texture features in the frequency domain, the current fault texture features in the time-frequency domain, the voltage fault texture features in the frequency domain, and the voltage fault texture features in the time-frequency domain, the sub-classifiers support the use of CNN and fully connected layers.

[0054] In a second aspect, the present invention provides a charging pile fault detection and diagnosis device based on a neural network, including: at least one processing unit, the processing unit is connected to a storage unit through a bus unit, the storage unit stores a computer program, and when the computer program is executed by the processing unit, the charging pile fault detection and diagnosis method based on the neural network is implemented.

[0055] In a third aspect, the present invention provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the charging pile fault detection and diagnosis method based on the neural network is implemented.

[0056] The above technical solutions provided by the embodiments of the present invention have the following advantages compared with the prior art:

[0057] The key point of the present invention is the charging pile fault detection and diagnosis method based on a neural network, which collects various fault characteristics of the charging pile; for various fault characteristics, corresponding sub-classifiers are constructed, and the sub-classifiers are weak classifiers for fault prediction based on the corresponding fault characteristics; evidence reasoning is performed, and the output results of each sub-classifier are fused to obtain the final fault diagnosis result that combines the prediction and diagnosis results of multiple sub-classifiers. By combining weak classifiers that analyze faults using different modalities of fault characteristics through evidence reasoning, evidence reasoning can well handle uncertainty and incomplete information. It can not only handle probability information, but also handle uncertainty more extensive than probability. Effectively merge information from different sub-classifiers, and can give a reasonable synthesis result even when there is a conflict between evidences, and has a certain robustness to the noise or errors of a single evidence source.

[0058] The sub-classifier of the present application uses two-step random mapping and sparse coding to enhance fault features, with fewer levels, low diagnostic calculation cost, fast speed, and strong real-time performance; and avoids the complex backpropagation process through random mapping and ridge regression, with low training calculation cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0061] Figure 1 is a flowchart of a method for detecting and diagnosing charging pile faults based on a neural network provided by an embodiment of the present invention;

[0062] Figure 2 is a flowchart for determining the weights of sub-classifiers provided by an embodiment of the present invention;

[0063] Figure 3 is a flowchart for performing evidence reasoning to fuse the output results of each sub-classifier to obtain the final fault diagnosis result that fuses the prediction and diagnosis results of multiple sub-classifiers;

[0064] Figure 4 is a flowchart for recursively fusing the basic probability mass and the uncertain part probability mass of two pieces of evidence provided by an embodiment of the present invention;

[0065] Figure 5 is a schematic diagram of a device for detecting and diagnosing charging pile faults based on a neural network provided by an embodiment of the present invention. Detailed implementation manners

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0067] It should be noted that in this article, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or elements inherent to such a process, method, article, or device. Without further limitations, the element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0068] Embodiment 1

[0069] To quickly diagnose charging pile fault problems, such as Figure 1As shown in the figure, the present invention provides a method for detecting and diagnosing charging pile faults based on a neural network. The algorithms used mainly include the following content:

[0070] Collect various fault features of the charging pile; the fault features are physical parameter features of the charging pile that can provide support for fault classification, including: current fault texture features, voltage fault texture features, and temperature features of the set components of the charging pile. In the specific implementation process, the current fault texture features can be any one or a combination of the following types: statistical current fault texture features: current average value, standard deviation, smoothness, skewness, kurtosis, and root mean square value; frequency-domain current fault texture features: current frequency-domain features after Fourier transform of current data; time-frequency domain current fault texture features: current time-frequency domain features obtained by wavelet transform of current data; the voltage fault texture features can be a combination of one or more of the following types: statistical voltage fault texture features: voltage average value, standard deviation, smoothness, skewness, kurtosis, and root mean square value; frequency-domain voltage fault texture features: voltage frequency-domain features after Fourier transform of voltage data; time-frequency domain voltage fault texture features: voltage time-frequency domain features obtained by wavelet transform of voltage data.

[0071] For various fault features, corresponding sub-classifiers are constructed, including:

[0072] It is assumed that there are a total of K types of fault features, and the k-th type of fault feature is X k , and the sub-classifier for the k-th type of fault feature X k is C k . As Figure 2 shown, the calculation method of the classifier weights of all sub-classifiers is as follows:

[0073] Generate feature nodes by randomly mapping and sparsely coding the fault features:

[0074]

[0075] Among them, is the m-th feature node corresponding to the sub-classifier C k , is the corresponding random weight matrix, is the corresponding random bias, a 1 () is a non-linear activation function, M is the total number of feature nodes, and K is the total number of sub-classifiers;

[0076] Concatenate all the feature nodes of each sub-classifier:

[0077]

[0078] ​​​It can be seen that each feature node corresponds to a random weight matrix and a random bias, constructing multiple groups of non - linear feature combinations based on fault features, and expanding the representation of fault features through the random mapping and sparse coding of the random weight matrix and the random bias.

[0079] Further random mapping and sparse coding are respectively performed on the concatenation results of all feature nodes of each sub - classifier to generate enhanced feature nodes:

[0080]

[0081] Among them, is the nth enhanced feature node corresponding to the sub - classifier C k , is the random weight matrix corresponding to , is the random bias corresponding to , a 2 () is a non - linear activation function, and N is the total number of enhanced feature nodes;

[0082] Concatenate all the enhanced feature nodes of each type of sub - classifier

[0083] Concatenate the concatenation results of all feature nodes and the concatenation results of enhanced feature nodes of each type of sub - classifier into the input matrix of each type of sub - classifier, and calculate the weights of each type of sub - classifier through ridge regression based on the input matrix and the charging pile fault type labels:

[0084]

[0085] Among them, W k is the weight of the kth sub - classifier, is the pseudo - inverse matrix of the input matrix of each type of sub - classifier, and Y is the charging pile fault type label.

[0086] In the specific implementation process, for the current fault texture features in the frequency domain, the current fault texture features in the time - frequency domain, the voltage fault texture features in the frequency domain, and the voltage fault texture features in the time - frequency domain, the sub - classifier supports the use of CNN and fully - connected layers. The fully - connected layer uses the sigmoid activation function to generate the probability of each type of charging pile fault type label according to the fault texture features.

[0087] Perform evidence reasoning, fuse the output results of each sub - classifier, and obtain the final fault diagnosis result that fuses the prediction and diagnosis results of multiple sub - classifiers. In the charging pile fault diagnosis, the outputs of multiple sub - classifiers are fused through evidence reasoning to improve the accuracy and robustness of the diagnosis. As Figure 3 shown, the following are the specific implementation steps and combination methods of evidence reasoning, including:

[0088] Determine the recognition range as Y, where Y is the set of charging pile fault type labels, Y = {Y 1 , Y 2 ..., Y i ,...Y I}, where Y i represents the label of the i-th charging pile fault type.

[0089] Each sub-classifier assigns the confidence of the predicted labels of various fault types for various fault features Among them, based on the sub-classifier weights W k , k = 1, 2,..., K and the input matrix of all charging pile fault types, the predictions of all charging pile fault types of each sub-classifier are obtained: The confidence is the predicted probability of each fault type obtained by the sub-classifier based on the fault features;

[0090] Calculate the global uncertainty based on the confidence

[0091]

[0092] Global uncertainty represents the uncertain part of the evidence; for example, if it is set that the temperature feature of a component can only be used for the fault diagnosis of this component and its associated components, and it is impossible to make a fault diagnosis for other parts that are not associated with this component, which has strong uncertainty.

[0093] Determine the accuracy of the predicted labels of various charging pile fault types of each sub-classifier, and assign the weights of the predicted labels of various charging pile fault types of the sub-classifier according to the accuracy

[0094]

[0095] Among them, acc() is the accuracy function, is the accuracy of the k-th sub-classifier predicting the label of the i-th charging pile fault type, is the accuracy of the l-th sub-classifier predicting the label of the i-th charging pile fault type.

[0096] The weights of the predicted labels of various charging pile fault types of each sub-classifier are weighted by the results of the predicted labels of various charging pile fault types of each sub-classifier to obtain the basic probability mass of the evidence for each fault category:

[0097]

[0098] Then the probability mass of the uncertain part is:

[0099]

[0100] The probability mass of the uncertain part consists of two parts: the unallocated probability due to weights and the original uncertain part of the evidence

[0101] such as Figure 4 shown, recursively fuse the basic probability mass and the probability mass of the uncertain part of two pieces of evidence: recursively combine the basic probability masses of two pieces of evidence into a new basic probability mass using the combination formula, and combine the probability masses of the uncertain parts of two pieces of evidence into a new probability mass of the uncertain part, repeat this process until the basic probability masses and the probability masses of the uncertain parts of all pieces of evidence are combined. The combination formula includes:

[0102] Recursively combine two basic probability masses:

[0103]

[0104] Recursively calculate the probability mass of the combined uncertain part:

[0105]

[0106] Finally,

[0107] where λ is the normalization factor, subtract the sum of the products of the basic probability masses of all conflict terms. All conflict terms are the combinations of all different charging pile fault types ij. The basic probability mass of each charging pile fault type after combination is The probability mass of the uncertain part of each charging pile fault type after combination is β Y . The meaning of the above combination formula is that starting from k = 2, recursively combine two basic probability masses according to After execution, k is incremented by one until k = K, which realizes recursively fusing the (k - 1)-th basic probability mass into the k-th basic probability mass; starting from k = 2, calculate the probability mass of the combined uncertain part recursively according to After execution, k is incremented by one until k = K, which realizes recursively fusing the (k - 1)-th probability mass of the uncertain part into the k-th probability mass of the uncertain part.

[0108] According to the basic probability mass of each charging pile fault type after combination The probability mass of the uncertain part of each charging pile fault type after combination β Y Calculate the confidence distribution after combination

[0109] According to the confidence distribution after synthesis Select the charging pile fault type as the final fault diagnosis result.

[0110] Embodiment 2

[0111] Refer to Figure 5 As shown, an embodiment of the present invention provides a charging pile fault detection and diagnosis device based on a neural network, including: at least one processing unit, the processing unit is connected to a storage unit through a bus unit, and the storage unit is used as a computer-readable storage medium and can be used to store software programs, computer-executable programs, and modules, such as the software programs, computer-executable programs, and modules corresponding to a charging pile fault detection and diagnosis method based on a neural network in an embodiment of the present invention. The processing unit realizes the above-mentioned charging pile fault detection and diagnosis method based on a neural network by running the software programs, computer-executable programs, and modules stored in the storage unit, including:

[0112] Collect various fault features of the charging pile; the fault features are physical parameter features of the charging pile that can provide support for fault classification, including: current fault texture features, voltage fault texture features, and temperature features of the set components of the charging pile;

[0113] For various fault features, construct corresponding sub-classifiers, and the sub-classifiers are weak classifiers for fault prediction based on the corresponding fault features;

[0114] Perform evidence reasoning, fuse the output results of each sub-classifier, and obtain the final fault diagnosis result that combines the prediction and diagnosis results of multiple sub-classifiers.

[0115] Certainly, the computer program stored in the storage unit of the charging pile fault detection and diagnosis device based on a neural network provided by the embodiment of the present invention is not limited to the method operations described above, and can also execute the related operations in a charging pile fault detection and diagnosis method based on a neural network provided by any embodiment of the present invention.

[0116] Embodiment 3

[0117] An embodiment of the present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the computer program is executed, it realizes the above-mentioned charging pile fault detection and diagnosis method based on a neural network, including:

[0118] Collect various fault features of the charging pile; the fault features are physical parameter features of the charging pile that can provide support for fault classification, including: current fault texture features, voltage fault texture features, and temperature features of the set components of the charging pile;

[0119] For various fault characteristics, corresponding sub-classifiers are constructed, and the sub-classifiers are weak classifiers for fault prediction based on the corresponding fault characteristics;

[0120] Evidence reasoning is carried out to fuse the output results of each sub-classifier to obtain the final fault diagnosis result that integrates the prediction and diagnosis results of multiple sub-classifiers.

[0121] The computer-readable storage medium provided by the embodiments of the present invention stores computer programs that are not limited to the method operations described above, and can also execute the related operations in a method for detecting and diagnosing charging pile faults based on a neural network provided by any embodiment of the present invention.

[0122] In the embodiments provided by the present invention, it should be understood that the disclosed structure and method can be implemented in other ways. For example, the structural embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of the structure or unit can be in an electrical, mechanical or other form.

[0123] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0124] In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0125] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A charging pile fault detection and diagnosis method based on neural network, characterized in that: include: Collect various fault characteristics of charging piles; The fault characteristics are physical parameter characteristics of the charging pile that can provide support for fault classification, including: current fault texture characteristics, voltage fault texture characteristics and charging pile setting component temperature characteristics; For each type of fault feature, a corresponding sub-classifier is constructed, wherein the sub-classifier is a weak classifier for fault prediction based on the corresponding fault feature; Perform evidence reasoning, merge the output results of each sub-classifier, and obtain the final fault diagnosis result that integrates the prediction and diagnosis results of multiple sub-classifiers.

2. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 1, characterized in that: The current fault texture feature adopts any one or a combination of the following types: Statistical current fault texture features: current mean, standard deviation, smoothness, skewness, kurtosis and RMS value; Frequency domain current fault texture features: the current frequency domain features after Fourier transform of current data; Current fault texture features in time-frequency domain: Current time-frequency domain features obtained by wavelet transform of current data; Voltage fault texture features are one or a combination of the following types: Statistical voltage fault texture features: voltage mean, standard deviation, smoothness, skewness, kurtosis and RMS value; Frequency domain voltage fault texture features: voltage frequency domain features after voltage data is transformed by Fourier transform; Voltage fault texture features in the time-frequency domain: voltage time-frequency domain features obtained by wavelet transform of voltage data.

3. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 1, characterized in that: According to various fault characteristics, corresponding sub-classifiers are constructed, including: Set the kth fault feature X k , for the k-th fault feature X k The subclassifier is C k , the classifier weights of all sub-classifiers are calculated as follows: Generate feature nodes by randomly mapping and sparsely encoding fault features: in, The corresponding sub-classifier is C k The mth feature node of For the corresponding The random weight matrix, For the corresponding The random bias of , a1() is the nonlinear activation function, M is the total number of feature nodes, and K is the total number of sub-classifiers; Concatenate all feature nodes of each sub-classifier: The concatenation results of all feature nodes of each sub-classifier are further randomly mapped and sparsely coded to generate enhanced feature nodes: in, The corresponding sub-classifier is C k The nth enhanced feature node, For the corresponding The random weight matrix, For the corresponding The random bias of , a2() is the nonlinear activation function, and N is the total number of enhanced feature nodes; Concatenate all enhanced feature nodes of various sub-classifiers All the feature node splicing results and enhanced feature node splicing results of various sub-classifiers are spliced ​​into the input matrix of various sub-classifiers, and the weights of various sub-classifiers are calculated based on the input matrix and the charging pile fault type label through ridge regression: Among them, W k is the weight of the kth sub-classifier, is the input matrix of each sub-classifier The pseudo-inverse matrix of , Y is the charging pile fault type label.

4. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 1, characterized in that: Perform evidence reasoning, merge the output results of each sub-classifier, and obtain the final fault diagnosis result that merges the prediction and diagnosis results of multiple sub-classifiers, including: Determine the identification range as Y, where Y is the charging pile fault type label set, Y = {Y1, Y2…, Y i ,…Y I }, where Y i Indicates the fault type label of the i-th charging pile; Each sub-classifier predicts the confidence of each fault type label based on the fault feature allocation Calculate global uncertainty based on confidence level Global uncertainty To indicate an uncertain part of the evidence; Assign weights to various charging pile fault types predicted by the sub-classifier according to the accuracy of the labels of various charging pile fault types predicted by the sub-classifier The weights of various charging pile fault types predicted by each sub-classifier are weighted. The results of each sub-classifier predicting various charging pile fault types are obtained by obtaining evidence. For the basic probability mass of each fault category, the probability mass of the uncertain part is calculated; Recursively merge the basic probability mass and the uncertain partial probability mass of two pieces of evidence; Calculate the synthesized confidence distribution Among them, the basic probability mass of each charging pile fault type after synthesis is The probability mass of the uncertain part of each charging pile fault type after synthesis is β Y ; According to the synthesized confidence distribution Select the charging pile fault type as the final fault diagnosis result.

5. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 4, characterized in that: Subclassifier weight W based on the current iteration step k ,k=1,2,...,K and the input matrix are used to calculate the predictions of all charging pile fault type labels for each sub-classifier Where I is the total number of charging pile fault type labels; determine the accuracy of each type of charging pile fault type label predicted by each sub-classifier, and allocate the contribution of each type of charging pile fault type predicted by the sub-classifier according to the accuracy of each type of charging pile fault type label predicted by the sub-classifier Among them, acc() is the accuracy function, The accuracy of the k-th sub-classifier in predicting the fault type label of the i-th charging pile.

6. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 4, characterized in that: The weights of the various charging pile fault types predicted by each sub-classifier are weighted. The results of each sub-classifier predicting the various charging pile fault types are obtained to obtain the basic probability mass of evidence for each fault category: The probability mass of the uncertain part is: The probability mass of the uncertain part consists of two parts Composition: Unassigned probabilities due to weights and the original uncertain parts of the evidence 7. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 4, characterized in that: Recursively fusing the basic probability masses of two pieces of evidence includes: recursively merging the basic probability masses of the two pieces of evidence into a new basic probability mass using a synthesis formula, merging the uncertain partial probability masses of the two pieces of evidence into a new uncertain partial probability mass, and repeating this process until the basic probability masses and the uncertain partial probability masses of all pieces of evidence are merged, wherein the synthesis formula includes: Recursively synthesize two basic probability masses: Recursively calculate the probability mass of the combined uncertain part: final, Among them, λ is the normalization factor, Subtract the sum of the products of the basic probability masses of all conflicting items, that is, the combination of all different charging pile fault types ij, and the basic probability mass of each charging pile fault type after synthesis is The probability mass of the uncertain part of each charging pile fault type after synthesis is β Y .

8. The method for detecting and diagnosing charging pile faults based on a neural network according to claim 1, characterized in that: For the current fault texture features of the frequency domain type, the current fault texture features of the time-frequency domain type, the voltage fault texture features of the frequency domain type, and the voltage fault texture features of the time-frequency domain type, the sub-classifier supports the use of CNN and fully connected layers.

9. A charging pile fault detection and diagnosis device based on a neural network, characterized in that: include: At least one processing unit, the processing unit is connected to a storage unit via a bus unit, the storage unit stores a computer program, and when the computer program is executed by the processing unit, the charging pile fault detection and diagnosis method based on a neural network as described in any of claims 1-8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the charging pile fault detection and diagnosis method based on neural network as described in any of claims 1-8 is implemented.

Images