DC fault arc detection and model training method, device, equipment and medium
By combining EMD and CNN recognition models for DC current signals, the problem of inaccurate arc detection in the prior art is solved, and higher detection accuracy and adaptability are achieved.
Patent Information
- Application Number
- CN202411871048.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-04-08
AI Technical Summary
The existing DC fault arc detection methods are not accurate enough in complex scenarios, difficult to adapt to multiple scenarios, and the detection rules cannot adapt to changes.
By performing empirical modal decomposition (EMD) of the DC signal, multi-layer eigenmodal function (IMF) components are extracted, N-layer IMF components with the largest amount of effective information are selected, and M time domain features are extracted, and input them into the pre-trained convolutional neural network (CNN) recognition model for detection.
It improves the accuracy and adaptability of DC fault arc detection, can extract features more effectively, adapt to complex scenarios, fast detection speed and high real-time performance.
Smart Images
Figure CN119782934A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of power fault detection, and in particular to a DC fault arc detection and model training method, device, equipment and medium. Background Art
[0002] With the widespread application and development of photovoltaic power generation systems, their safety issues are becoming increasingly apparent. During the operation of the power station, if the connector is not installed to professional standards, is subjected to natural weathering for a long time, or is bitten by animals, it may induce the occurrence of fault arc phenomena. If the fault arc is not handled quickly and properly, its continuous arcing state will cause extremely high temperatures, which can easily ignite the surrounding combustible materials and cause fire accidents. Therefore, accurate detection of fault arcs is crucial.
[0003] At present, the schemes for DC fault arc detection in photovoltaic power generation systems are mainly based on the detection methods of the physical characteristics of the arc (such as arc light, arc sound, heat, electromagnetic radiation characteristics, etc.) and the detection methods based on the time-frequency domain characteristics of voltage and current. Since the physical characteristics of the arc are greatly affected by the interference of the complex environment and are restricted and limited by the sensors, the detection of fault arcs based on this is not accurate enough; therefore, the detection method based on the time-frequency domain characteristics of arc current and voltage is currently the most widely used detection method, but in photovoltaic power generation systems, the scenes are complex and the components are numerous. Normal switching actions and load mutations may lead to misjudgments, and the detection rules cannot change adaptively with the scenes. Therefore, the existing detection methods cannot quickly, stably and reliably detect DC fault arcs, and are difficult to adapt to complex and diverse scenes. Summary of the invention
[0004] In order to solve the problems in the related art, the embodiments of the present disclosure provide a method, device, equipment and medium for DC fault arc detection and model training thereof.
[0005] In a first aspect, an embodiment of the present disclosure provides a DC fault arc detection method, comprising:
[0006] Acquire a DC signal;
[0007] Performing empirical mode decomposition (EMD) on the DC signal to obtain multi-layer intrinsic mode function (IMF) components;
[0008] Select N layers of IMF components with the largest effective information amount from the multi-layer IMF components, where N is an integer greater than or equal to 2;
[0009] Extracting M time domain features of each layer of IMF components in the N layers of IMF components to obtain N*M features, where M is an integer greater than or equal to 1;
[0010] The N*M features are input into a pre-trained convolutional neural network (CNN) recognition model to obtain a DC fault arc detection result output by the pre-trained CNN recognition model.
[0011] In a possible implementation manner, selecting N layers of IMF components with the largest amount of effective information from the multiple layers of IMF components includes:
[0012] Calculating the information entropy of each layer of IMF components in the multiple layers of IMF components and the information entropy of the direct current signal;
[0013] Calculating the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the direct current signal;
[0014] The N-layer IMF components with the largest ratio among the multiple layers of IMF components are obtained as the N-layer IMF components with the largest effective information amount.
[0015] In a possible implementation, the M time domain features of each layer of IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of IMF components.
[0016] In a possible implementation, the method further includes:
[0017] Acquire a sample direct current signal and a label corresponding to the signal, wherein the label includes a fault arc or a non-fault arc;
[0018] Performing EMD on the sample DC signal to obtain multi-layer sample IMF components;
[0019] Selecting N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components;
[0020] Extracting M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features;
[0021] Based on N*M sample features of multiple sample direct current signals and their corresponding labels, a CNN recognition model is trained to obtain a trained CNN recognition model.
[0022] In a second aspect, the present disclosure provides a DC fault arc detection model training method, including:
[0023] Acquire a sample direct current signal and a label corresponding to the signal, wherein the label includes a fault arc or a non-fault arc;
[0024] Performing empirical mode decomposition (EMD) on the sample direct current signal to obtain multi-layer sample intrinsic mode function (IMF) components;
[0025] Select the N-layer sample IMF components with the largest amount of information from the multi-layer sample IMF components;
[0026] Extracting M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features;
[0027] Based on N*M sample features of multiple sample direct current signals and their corresponding labels, a CNN recognition model is trained to obtain a trained CNN recognition model.
[0028] In a possible implementation manner, selecting N layers of sample IMF components with the largest amount of information from the multiple layers of sample IMF components includes:
[0029] Calculating the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal;
[0030] Calculating the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal;
[0031] The N-layer sample IMF components with the largest ratio among the multi-layer sample IMF components are obtained as the N-layer sample IMF components with the largest effective information amount.
[0032] In a third aspect, an embodiment of the present disclosure provides a DC fault arc detection device, comprising:
[0033] A signal acquisition module is configured to acquire a direct current signal;
[0034] A mode decomposition module is configured to perform empirical mode decomposition (EMD) on the DC signal to obtain multi-layer intrinsic mode function (IMF) components;
[0035] A component selection module is configured to select N layers of IMF components with the largest effective information amount from the multiple layers of IMF components, where N is an integer greater than or equal to 2;
[0036] A feature extraction module is configured to extract M time domain features of each layer of IMF components in the N layers of IMF components to obtain N*M features, where M is an integer greater than or equal to 1;
[0037] The recognition module is configured to input the N*M features into a pre-trained convolutional neural network (CNN) recognition model to obtain a DC fault arc detection result output by the pre-trained CNN recognition model.
[0038] In a possible implementation manner, the component selection module is configured as follows:
[0039] Calculating the information entropy of each layer of IMF components in the multiple layers of IMF components and the information entropy of the direct current signal;
[0040] Calculating the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the direct current signal;
[0041] The N-layer IMF components with the largest ratio among the multiple layers of IMF components are obtained as the N-layer IMF components with the largest effective information amount.
[0042] In a possible implementation, the M time domain features of each layer of IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of IMF components.
[0043] In a possible implementation, the device further includes:
[0044] The training module is configured to obtain a sample DC signal and a corresponding label thereof, wherein the label includes a faulty arc or a non-faulty arc; perform EMD on the sample DC signal to obtain a multi-layer sample IMF component; select N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components; extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features; train a CNN recognition model based on the N*M sample features of multiple sample DC signals and their corresponding labels to obtain a trained CNN recognition model.
[0045] In a fourth aspect, an embodiment of the present disclosure provides a DC fault arc detection model training device, comprising:
[0046] A sample acquisition module is configured to acquire a sample DC signal and a label corresponding thereto, wherein the label includes a fault arc or a non-fault arc;
[0047] A sample mode decomposition module is configured to perform empirical mode decomposition (EMD) on the sample DC signal to obtain multi-layer sample intrinsic mode function (IMF) components;
[0048] A sample component selection module is configured to select N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components;
[0049] A sample feature extraction module is configured to extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features;
[0050] The model training module is configured to train the CNN recognition model based on N*M sample features of multiple sample DC signals and their corresponding labels to obtain a trained CNN recognition model.
[0051] In a possible implementation, the sample component selection module is configured as follows:
[0052] Calculating the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal;
[0053] Calculating the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal;
[0054] The N-layer sample IMF components with the largest ratio among the multi-layer sample IMF components are obtained as the N-layer sample IMF components with the largest effective information amount.
[0055] In a fifth aspect, an embodiment of the present disclosure provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, and wherein the one or more computer instructions are executed by the processor to implement a method as described in any one of the first aspect or the second aspect.
[0056] In a sixth aspect, a computer-readable storage medium is provided in an embodiment of the present disclosure, on which computer instructions are stored. When the computer instructions are executed by a processor, a method as described in any one of the first aspect or the second aspect is implemented.
[0057] According to the technical solution provided by the embodiment of the present disclosure, after obtaining the DC signal, the DC signal can be subjected to EMD to obtain multiple layers of IMF components. Since too many IMF components are prone to redundancy, N layers of IMF components with the largest amount of effective information can be selected from the multiple layers of IMF components, and M time domain features of each layer of IMF components in the N layers of IMF components can be extracted to obtain N*M features; the N*M features are input into a pre-trained convolutional neural network CNN recognition model to obtain the DC fault arc detection result output by the pre-trained CNN recognition model. Since the IMF components contained in each electrical appliance in the power system are different, the present embodiment can adaptively select the N layers of IMF components with the largest amount of effective information, which can solve the situation where there are many components and different information in complex scenes. By adaptively selecting N layers of IMF components, features can be extracted more effectively, the detection capability and detection accuracy of DC fault arcs can be improved, and various complex scenes can be adapted. Moreover, the present embodiment uses EMD and CNN recognition models to detect DC fault arcs, and the solution is simple, the detection speed is faster, and the real-time performance is high.
[0058] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] Other features, objectives and advantages of the present disclosure will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings:
[0060] Figure 1 A flow chart of a DC fault arc detection method provided by an embodiment of the present disclosure is shown.
[0061] Figure 2 A flow chart of a DC fault arc detection model training method provided according to an embodiment of the present disclosure is shown.
[0062] Figure 3 A structural block diagram of a DC fault arc detection device provided by an embodiment of the present disclosure is shown.
[0063] Figure 4 A structural block diagram of a DC fault arc detection model training device provided by an embodiment of the present disclosure is shown.
[0064] Figure 5 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.
[0065] Figure 6 A schematic diagram showing the structure of a computer system suitable for implementing the method of the embodiment of the present disclosure is shown. DETAILED DESCRIPTION
[0066] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.
[0067] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or a combination thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or a combination thereof exist or are added.
[0068] It should also be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0069] Figure 1 FIG. 2 is a flow chart of a DC fault arc detection method provided by an embodiment of the present disclosure. Figure 1 As shown, the DC fault arc detection method includes the following steps S101-S105:
[0070] In step S101, a direct current signal is obtained;
[0071] In step S102, the direct current signal is subjected to empirical mode decomposition (EMD) to obtain multi-layer intrinsic mode function (IMF) components;
[0072] In step S103, N layers of IMF components with the largest amount of information are selected from the multiple layers of IMF components, where N is an integer greater than or equal to 2;
[0073] In step S104, M time domain features of each layer of IMF components in the N layers of IMF components are extracted to obtain N*M features;
[0074] In step S105, the N*M features are input into a pre-trained convolutional neural network (CNN) recognition model to obtain a DC fault arc detection result output by the pre-trained CNN recognition model.
[0075] In a possible implementation manner, the DC fault arc detection method is applied to an electronic device that can detect a DC fault arc in a DC power system.
[0076] In a possible implementation, a direct current signal in a direct current power system may be collected through a transformer. The direct current signal may be a direct current voltage signal or a direct current signal, which is not limited here.
[0077] In one possible implementation, Empirical Mode Decomposition (EMD) is an adaptive signal time-frequency processing method suitable for processing nonlinear and non-stationary time series signals. The core idea is to use the characteristic time scale to determine the modal function of the effective signal in the signal, and to adaptively decompose the original signal into a series of orthogonal intrinsic mode function (IMF) components. Each IMF component represents a frequency component of the signal, has the stationary signal characteristics and physical meaning of a single component, and can better describe the essential characteristics of the signal. The DC signal in this implementation is a time series signal formed by electrical signals collected in time sequence. By performing EMD on the DC signal, multiple layers of IMF components can be obtained, and each layer of IMF components represents a frequency component of the DC signal.
[0078] In a possible implementation manner, performing EMD on the DC signal may include the following steps:
[0079] Step 1: Obtain all local extreme points of the DC signal and use the cubic spline function to calculate the upper envelope u of the local extreme points1 (t) and the lower envelope u 2 (t) find its local mean m(t):
[0080]
[0081] Step 2: Calculate h according to the following formula 1k (t):
[0082] h 1k (t) = h 1(k-1) (t)-m 1k (t);
[0083] In the above formula, when k = 1, h 10 (t) is the DC signal x(t), m 11 (t) is m(t) calculated in step 1. When k>1, m 1k (t) is h 1(k-1) (t); repeat the calculation k times. Since too many repetitions will cause the IMF component to become a frequency modulated signal with a fixed amplitude, the standard deviation can be used as the basis for judging the end of the decomposition process. The standard deviation expression is: Where T is h 1(k-1) The total number of discrete signals in (t), when S D When convergence occurs, the calculation is no longer repeated. At this time, the first IMF component C of the DC signal x(t) is obtained. 1 =h 1k (t) and the residual term r 1 (t), the residual term r 1 (t) = x(t) - C 1 .
[0084] Step 3: Add r 1 (t) Continue to decompose according to the above steps 1 and 2 to find C 2 , C 3 When it is determined that the residual term is monotonic or the absolute value of the residual term is extremely small, such as less than a predetermined threshold, the decomposition step stops, and the DC signal x(t) is decomposed by EMD into the following formula: Where N is the number of decomposition layers; C i (t) is the IMF component of the i-th layer; r N (t) is the residual term.
[0085] According to the above EMD decomposition steps, EMD can gradually extract the scale components that make up the DC signal from high to low frequency, thereby obtaining a series of IMF components and a residual remainder. Since high-frequency components usually represent the main characteristics and energy of the signal, the EMD method provided in this embodiment adopts a main component cyclic extraction strategy, that is, first extracting the high-frequency main components, and then gradually extracting the low-frequency partial components, so as to effectively reduce the signal feature loss, retain the local characteristics and time-frequency changes of the signal, and improve the accuracy and stability of signal decomposition.
[0086] In a possible implementation, the information characteristics of the IMF components decomposed from the DC signal are an efficient means of distinguishing the categories of DC signals, but at the same time, useless information will also increase relatively. Therefore, the effective information amount of each layer of IMF components in the multi-layer IMF components can be calculated, and the N-layer IMF components with the largest effective information amount are selected. For example, N can be 2, 3 or 4, and preferably, the N value can be 3.
[0087] In a possible implementation, for N layers of IMF components with the largest amount of effective information, M time domain features of each layer of IMF components can be extracted. The time domain features refer to the properties and statistical characteristics of the signal on the time axis, which describe the behavior and characteristics of the signal changing over time; in this way, N*M features can be obtained.
[0088] In a possible implementation, the neural network algorithms currently used for arc fault detection solutions are still mainly concentrated in relatively simple fully-linked neural network algorithms. Although fully-linked neural network algorithms are relatively easy to implement, there are problems such as difficulty in training and easy overfitting when there are too many parameters. The local perception and weight sharing of convolutional neural networks (CNN) can solve this problem, and CNN has good fault tolerance, parallel processing capabilities and self-learning capabilities, which can improve the accuracy of the detection algorithm compared to traditional fully-linked neural networks. Therefore, the implementation method uses a CNN recognition model to perform DC fault arc detection.
[0089] In one possible implementation, a CNN recognition model may be pre-trained. The CNN recognition model may learn the time domain characteristics of the first N layers of IMF components with a large amount of effective information in a DC signal with a historical DC fault arc and / or the time domain characteristics of the first N layers of IMF components with a large amount of effective information in a normal DC signal, and analyze the time domain characteristics of the first N layers of IMF components with a large amount of effective information in a currently acquired DC signal to obtain a DC fault arc detection result, which includes the presence of a DC fault arc or the absence of a DC fault arc.
[0090] After acquiring the DC signal, this embodiment can perform EMD on the DC signal to obtain multiple layers of IMF components. Since too many IMF components are prone to redundancy, N layers of IMF components with the largest amount of effective information can be selected from the multiple layers of IMF components, and M time domain features of each layer of IMF components in the N layers of IMF components can be extracted to obtain N*M features; the N*M features are input into a pre-trained convolutional neural network CNN recognition model to obtain the DC fault arc detection result output by the pre-trained CNN recognition model. Since each electrical appliance in the power system contains different IMF components, this embodiment can adaptively select the N layers of IMF components with the largest amount of effective information, which can solve the situation where there are many components and different information in complex scenes. By adaptively selecting N layers of IMF components, features can be extracted more effectively, the detection capability and detection accuracy of DC fault arcs can be improved, and various complex scenes can be adapted. In addition, this embodiment uses EMD and CNN recognition models to perform DC fault arc detection, which has a simple solution, faster detection speed, and high real-time performance.
[0091] In a possible implementation manner, selecting N layers of IMF components with the largest amount of effective information from the multiple layers of IMF components includes:
[0092] Calculating the information entropy of each layer of IMF components in the multiple layers of IMF components and the information entropy of the direct current signal;
[0093] Calculating the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the direct current signal;
[0094] The N-layer IMF components with the largest ratio among the multiple layers of IMF components are obtained as the N-layer IMF components with the largest effective information amount.
[0095] In this implementation, the information entropy measures the uncertainty or randomness of the information content. The higher the information entropy, the more complex and uncertain the information it contains. The information entropy of each layer of IMF components in the multi-layer IMF components and the information entropy of the DC signal can be calculated, and the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the DC signal is calculated. The larger the ratio, the greater the distinction between the information in this layer of IMF components and the DC signal, and the more obvious the effective information contained. Finally, the N layers of IMF components with the largest ratio are used as the N layers of IMF components to be selected.
[0096] In a possible implementation, the M time domain features of each layer of IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of IMF components.
[0097] In this implementation, the j-th layer IMF component can be denoted as C j (t), each layer of IMF component is a time series discrete signal, which can be recorded as x i is the amplitude of the ith discrete signal in the jth layer IMF component, and there are N discrete signals in the jth layer IMF component. The above time domain features are explained below.
[0098] The variance is a statistic that measures the degree of dispersion of each data in the signal from the mean, which can be expressed as follows:
[0099]
[0100] In the formula, x var The square sum of the difference between the signal amplitude and the mean value; F m is the signal average value.
[0101] The mean is calculated as shown in the following formula:
[0102] In the formula, x sum is the sum of the signal amplitudes.
[0103] The mean square value is x i The mean of the squares of is given by the following formula:
[0104] In the formula, x msv is the sum of the squares of the signal amplitudes.
[0105] The rectified average is expressed as the average of the absolute values of the signals and is calculated as follows:
[0106] In the formula, |x i | is the amplitude of the ith discrete signal x i The absolute value of .
[0107] The kurtosis reflects the numerical statistics of the random variable distribution characteristics of the IMF component of the DC signal. As the arc fault occurs and the severity increases, the kurtosis of the IMF component of the DC signal increases, indicating that the arc fault is more serious and it is easier to extract the arc fault information. The calculation formula of the kurtosis is as follows:
[0108]
[0109] The skewness indicates the degree to which the center of the probability density function of the IMF component of the DC signal deviates from the standard normal distribution, which can reflect the asymmetry of the amplitude distribution of the IMF component of the DC signal compared with the average value. Its calculation formula is as follows:
[0110] In the formula, xavg is the signal mean of the IMF component; σ is the standard deviation of the signal of the IMF component.
[0111] The margin factor is defined as the ratio of the signal peak value to the square root amplitude, which can be used to indicate the fullness of the waveform. The calculation formula is as follows:
[0112] In the formula, x max The signal x representing the IMF component i The maximum value in x sra Represents the square root amplitude.
[0113] The waveform factor is defined as the ratio of the effective value of the signal to the rectified average value, which can be used to indicate the degree of difference between the waveform and the standard sine wave. The calculation formula is as follows:
[0114]
[0115] The kurtosis factor is very sensitive to the impact component in the signal. In arc detection, the kurtosis factor can measure the smoothness of the arc current pulse. The calculation formula is as follows:
[0116] Where β is the signal kurtosis value; F c is the effective value of the signal.
[0117] The pulse factor is defined as the ratio of the signal peak value to the rectified average value, which can be used to represent the impact nature of the signal. The calculation formula is as follows:
[0118] In the formula, x max The signal x representing the IMF component i The maximum value in x arv is the rectified average of the signal of the IMF component.
[0119] In a possible implementation, the method further includes:
[0120] Acquire a sample direct current signal and a label corresponding to the signal, wherein the label includes a fault arc or a non-fault arc;
[0121] Performing EMD on the sample DC signal to obtain multi-layer sample IMF components;
[0122] Selecting N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components;
[0123] Extracting M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features;
[0124] Based on N*M sample features of multiple sample direct current signals and their corresponding labels, a CNN recognition model is trained to obtain a trained CNN recognition model.
[0125] In this embodiment, the electronic device may also pre-train a CNN recognition model, obtain historically collected sample DC signals and their corresponding labels as a training data set, wherein the labels include a faulty arc or a non-faulty arc.
[0126] In this embodiment, the sample DC signal can be subjected to empirical mode decomposition (EMD) to obtain multi-layer sample IMF components, and N layers of sample IMF components with the largest amount of information can be selected from the multi-layer sample IMF components. The selection scheme can be to calculate the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal; calculate the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal; obtain the N layers of sample IMF components with the largest ratio in the multi-layer sample IMF components as the N layers of sample IMF components with the largest amount of information; similarly, extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features; then, the initial CNN recognition model can be trained based on the N*M sample features of the sample DC signal and their corresponding labels to obtain a trained CNN recognition model.
[0127] The present disclosure also provides a DC fault arc detection model training method, Figure 2 FIG. 1 is a flow chart of a DC fault arc detection model training method provided according to an embodiment of the present disclosure. Figure 2 As shown, the DC fault arc detection model training includes the following steps S201-S205:
[0128] In step S201, a sample DC signal and its corresponding label are obtained, wherein the label includes a fault arc or a non-fault arc;
[0129] In step S202, performing empirical mode decomposition (EMD) on the sample DC signal to obtain multi-layer sample IMF components;
[0130] In step S203, N layers of sample IMF components with the largest amount of information are selected from the multi-layer sample IMF components;
[0131] In step S204, M time domain features of each layer of sample IMF components in the N layers of sample IMF components are extracted to obtain N*M sample features;
[0132] In step S205, a CNN recognition model is trained based on the N*M sample features of the sample DC power signal and their corresponding labels to obtain a trained CNN recognition model.
[0133] In a possible implementation manner, the DC fault arc detection model training method is applied to an electronic device that can train a DC fault arc detection model in a DC power system.
[0134] In a possible implementation, a DC signal in a DC power system can be collected by a transformer as a sample DC signal, and the DC signal can be a DC voltage signal or a DC current signal, which is not limited here. At the same time, a label corresponding to the sample DC signal is obtained, and the label is whether the sample DC signal is a DC signal when there is a fault arc or a normal DC signal when there is no fault arc.
[0135] In one possible implementation, empirical mode decomposition is an adaptive signal time-frequency processing method suitable for processing nonlinear and non-stationary time series signals. The core idea is to use the characteristic time scale to determine the modal function of the effective signal in the signal, and to adaptively decompose the original signal into a series of orthogonal intrinsic mode function (IMF) components. Each IMF component represents a frequency component of the signal, has the stationary signal characteristics and physical meaning of a single component, and can better describe the essential characteristics of the signal. The sample DC signal in this implementation is a time series signal formed by electrical signals collected in time sequence. By performing EMD on the sample DC signal, multiple layers of sample IMF components can be obtained, and each layer of sample IMF components represents a frequency component of the sample DC signal. The specific decomposition process can be referred to above and will not be repeated here.
[0136] In a possible implementation, for N layers of sample IMF components with the largest amount of effective information, M time domain features of each layer of sample IMF components can be extracted. The time domain features refer to the properties and statistical characteristics of the signal on the time axis, which describe the behavior and characteristics of the signal changing over time; in this way, N*M features can be obtained.
[0137] In a possible implementation, the M time domain features of each layer of sample IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of sample IMF components. The specific calculation process of these time domain features can be referred to above and will not be repeated here.
[0138] In a possible implementation, the neural network algorithms currently used for arc fault detection solutions are still mainly concentrated in relatively simple fully-linked neural network algorithms. Although fully-linked neural network algorithms are relatively easy to implement, there are problems such as difficulty in training and easy overfitting when there are too many parameters. The local perception and weight sharing of convolutional neural networks (CNN) can solve this problem, and CNN has good fault tolerance, parallel processing capabilities and self-learning capabilities, which can improve the accuracy of the detection algorithm compared to traditional fully-linked neural networks. Therefore, the implementation method uses a CNN recognition model to perform DC fault arc detection.
[0139] In a possible implementation, multiple sample DC signals can be obtained, and for each sample DC signal, its corresponding N*M features and labels can be obtained, and the N*M features are input into the CNN recognition model to obtain the DC fault arc detection result output by the CNN model, and based on the label corresponding to the sample DC signal, the accuracy of the DC fault arc detection result output by the CNN recognition model is calculated, and the model parameters of the CNN recognition model are continuously adjusted until the accuracy of the CNN recognition model reaches a predetermined threshold such as 97%, and then the training is stopped to obtain the trained CNN recognition model.
[0140] The CNN recognition model trained in this embodiment needs to adaptively select N layers of IMF components with the largest amount of effective information to solve the situation where there are many components and the information contained is different in complex scenarios. Through this adaptive selection of N layers of IMF components, features can be extracted more effectively, the detection capability and detection accuracy of DC fault arcs can be improved, and various complex scenarios can be adapted. In addition, this embodiment uses EMD and CNN recognition models to perform DC fault arc detection, which has a simple solution, faster detection speed, and high real-time performance.
[0141] In a possible implementation manner, selecting N layers of sample IMF components with the largest amount of information from the multiple layers of sample IMF components includes:
[0142] Calculating the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal;
[0143] Calculating the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal;
[0144] The N-layer sample IMF components with the largest ratio among the multi-layer sample IMF components are obtained as the N-layer sample IMF components with the largest effective information amount.
[0145] In this implementation, the information entropy measures the uncertainty or randomness of the information content. The higher the information entropy, the more complex and uncertain the information it contains. The information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal can be calculated, and the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal is calculated. The larger the ratio, the greater the distinction between the information in this layer of sample IMF components and the sample DC signal, and the more obvious the effective information contained. Finally, the N layers of sample IMF components with the largest ratio are used as the N layers of sample IMF components to be selected.
[0146] The present disclosure also provides a DC fault arc detection device, Figure 3 FIG. 1 is a block diagram of a DC fault arc detection device provided by an embodiment of the present disclosure. The device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 3 As shown, the DC fault arc detection device comprises:
[0147] The signal acquisition module 301 is configured to acquire a direct current signal;
[0148] The modal decomposition module 302 is configured to perform empirical mode decomposition (EMD) on the DC signal to obtain multi-layer intrinsic mode function (IMF) components;
[0149] The component selection module 303 is configured to select N layers of IMF components with the largest effective information amount from the multiple layers of IMF components, where N is an integer greater than or equal to 2;
[0150] A feature extraction module 304 is configured to extract M time domain features of each layer of IMF components in the N layers of IMF components to obtain N*M features, where M is an integer greater than or equal to 1;
[0151] The recognition module 305 is configured to input the N*M features into a pre-trained convolutional neural network (CNN) recognition model to obtain a DC fault arc detection result output by the pre-trained CNN recognition model.
[0152] In a possible implementation manner, the component selection module is configured as follows:
[0153] Calculating the information entropy of each layer of IMF components in the multiple layers of IMF components and the information entropy of the direct current signal;
[0154] Calculating the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the direct current signal;
[0155] The N-layer IMF components with the largest ratio among the multiple layers of IMF components are obtained as the N-layer IMF components with the largest effective information amount.
[0156] In a possible implementation, the M time domain features of each layer of IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of IMF components.
[0157] In a possible implementation, the device further includes:
[0158] The training module is configured to obtain a sample DC signal and a corresponding label thereof, wherein the label includes a faulty arc or a non-faulty arc; perform EMD on the sample DC signal to obtain a multi-layer sample IMF component; select N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components; extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features; train a CNN recognition model based on the N*M sample features of multiple sample DC signals and their corresponding labels to obtain a trained CNN recognition model.
[0159] The present disclosure also provides a DC fault arc detection model training device, Figure 4 The structure block diagram of a DC fault arc detection model training device provided by the embodiment of the present disclosure is shown. The device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 4 As shown, the DC fault arc detection model training device includes:
[0160] The sample acquisition module 401 is configured to acquire a sample DC signal and a label corresponding thereto, wherein the label includes a fault arc or a non-fault arc;
[0161] The sample mode decomposition module 402 is configured to perform empirical mode decomposition (EMD) on the sample DC signal to obtain multi-layer sample intrinsic mode function (IMF) components;
[0162] The sample component selection module 403 is configured to select N layers of sample IMF components with the largest amount of information from the multiple layers of sample IMF components;
[0163] The sample feature extraction module 404 is configured to extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features;
[0164] The model training module 405 is configured to train the CNN recognition model based on N*M sample features of multiple sample DC signals and their corresponding labels to obtain a trained CNN recognition model.
[0165] In a possible implementation, the sample component selection module is configured as follows:
[0166] Calculating the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal;
[0167] Calculating the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal;
[0168] The N-layer sample IMF components with the largest ratio among the multi-layer sample IMF components are obtained as the N-layer sample IMF components with the largest effective information amount.
[0169] The technical terms and technical features mentioned in the implementation manner of the present device are the same as or similar to those mentioned in the implementation manner of the above method. For the explanation and description of the technical terms and technical features involved in the present device, reference may be made to the explanation and description of the implementation manner of the above method, and they will not be repeated here.
[0170] The present disclosure also discloses an electronic device, Figure 5 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.
[0171] like Figure 5 As shown, the electronic device 500 includes a memory 501 and a processor 502, wherein the memory 501 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 502 to implement the method according to an embodiment of the present disclosure.
[0172] Figure 6 A schematic diagram showing the structure of a computer system suitable for implementing the method of the embodiment of the present disclosure is shown.
[0173] like Figure 6 As shown, the computer system 600 includes a processing unit 601, which can perform various processes in the above-mentioned embodiments according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the computer system 600 are also stored. The processing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0174] The following components are connected to the I / O interface 605: an input part 606 including a keyboard, a mouse, etc.; an output part 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage part 608 including a hard disk, etc.; and a communication part 609 including a network interface card such as a LAN card, a modem, etc. The communication part 609 performs communication processing via a network such as the Internet. The drive 610 is also connected to the I / O interface 605 as needed. Removable media 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., are installed on the drive 610 as needed, so that the computer program read therefrom is installed into the storage part 608 as needed. Among them, the processing unit 601 can be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, an NPU, etc.
[0175] In particular, according to an embodiment of the present disclosure, the method described above can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes computer instructions, and the computer instructions are executed by a processor to implement the method steps described above. In such an embodiment, the computer program product can be downloaded and installed from a network through the communication part 609, and / or installed from a removable medium 611.
[0176] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0177] The units or modules involved in the embodiments described in the present disclosure may be implemented by software or programmable hardware. The units or modules described may also be set in a processor, and the names of these units or modules do not constitute limitations on the units or modules themselves in some cases.
[0178] As another aspect, the present disclosure further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or computer system in the above embodiment; or a computer-readable storage medium that exists independently and is not assembled into a device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the method described in the present disclosure.
[0179] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other.
Claims
1. A DC fault arc detection method, characterized in that: include: Acquire a DC signal; Performing empirical mode decomposition (EMD) on the DC signal to obtain multi-layer intrinsic mode function (IMF) components; Select N layers of IMF components with the largest effective information amount from the multi-layer IMF components, where N is an integer greater than or equal to 2; Extracting M time domain features of each layer of IMF components in the N layers of IMF components to obtain N*M features, where M is an integer greater than or equal to 1; The N*M features are input into a pre-trained convolutional neural network (CNN) recognition model to obtain a DC fault arc detection result output by the pre-trained CNN recognition model.
2. The method according to claim 1, characterized in that The step of selecting N layers of IMF components with the largest amount of effective information from the multiple layers of IMF components includes: Calculating the information entropy of each layer of IMF components in the multiple layers of IMF components and the information entropy of the direct current signal; Calculating the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the direct current signal; The N-layer IMF components with the largest ratio among the multiple layers of IMF components are obtained as the N-layer IMF components with the largest effective information amount.
3. The method according to claim 1, characterized in that The M time domain features of each layer of IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of IMF components.
4. The method according to claim 1, characterized in that: The method further comprises: Acquire a sample direct current signal and a label corresponding to the signal, wherein the label includes a fault arc or a non-fault arc; Performing EMD on the sample DC signal to obtain multi-layer sample IMF components; Selecting N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components; Extracting M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features; Based on N*M sample features of multiple sample direct current signals and their corresponding labels, a CNN recognition model is trained to obtain a trained CNN recognition model.
5. A DC fault arc detection model training method, characterized in that: include: Acquire a sample direct current signal and a label corresponding to the signal, wherein the label includes a fault arc or a non-fault arc; Performing empirical mode decomposition (EMD) on the sample direct current signal to obtain multi-layer sample intrinsic mode function (IMF) components; Select N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components, where N is an integer greater than or equal to 2; Extracting M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features, where M is an integer greater than or equal to 1; Based on N*M sample features of multiple sample direct current signals and their corresponding labels, a convolutional neural network (CNN) recognition model is trained to obtain a trained CNN recognition model.
6. The method according to claim 5, characterized in that The step of selecting N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components includes: Calculating the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal; Calculating the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal; The N-layer sample IMF components with the largest ratio among the multi-layer sample IMF components are obtained as the N-layer sample IMF components with the largest effective information amount.
7. A DC fault arc detection device, characterized in that: include: A signal acquisition module is configured to acquire a direct current signal; A mode decomposition module is configured to perform empirical mode decomposition (EMD) on the DC signal to obtain multi-layer intrinsic mode function (IMF) components; A component selection module is configured to select N layers of IMF components with the largest effective information amount from the multiple layers of IMF components, where N is an integer greater than or equal to 2; A feature extraction module is configured to extract M time domain features of each layer of IMF components in the N layers of IMF components to obtain N*M features, where M is an integer greater than or equal to 1; The recognition module is configured to input the N*M features into a pre-trained convolutional neural network (CNN) recognition model to obtain a DC fault arc detection result output by the pre-trained CNN recognition model.
8. The device according to claim 7, characterized in that The component selection module is configured as follows: Calculating the information entropy of each layer of IMF components in the multiple layers of IMF components and the information entropy of the direct current signal; Calculating the ratio of the information entropy of each layer of IMF components in the multi-layer IMF components to the information entropy of the direct current signal; The N-layer IMF components with the largest ratio among the multiple layers of IMF components are obtained as the N-layer IMF components with the largest effective information amount.
9. The device according to claim 7, characterized in that The M time domain features of each layer of IMF components include at least one of the following features: variance, mean, mean square error, rectified mean, kurtosis, skewness, margin factor, waveform factor, kurtosis factor and pulse factor of each layer of IMF components.
10. The device according to claim 7, characterized in that The device also includes: The training module is configured to obtain a sample DC signal and a corresponding label thereof, wherein the label includes a faulty arc or a non-faulty arc; perform EMD on the sample DC signal to obtain a multi-layer sample IMF component; select N layers of sample IMF components with the largest amount of information from the multi-layer sample IMF components; extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features; train a CNN recognition model based on the N*M sample features of multiple sample DC signals and their corresponding labels to obtain a trained CNN recognition model.
11. A DC fault arc detection model training device, characterized in that: include: A sample acquisition module is configured to acquire a sample DC signal and a label corresponding thereto, wherein the label includes a fault arc or a non-fault arc; A sample mode decomposition module is configured to perform empirical mode decomposition (EMD) on the sample DC signal to obtain multi-layer sample intrinsic mode function (IMF) components; A sample component selection module is configured to select N layers of sample IMF components with the largest amount of information from the multiple layers of sample IMF components, where N is an integer greater than or equal to 2; A sample feature extraction module is configured to extract M time domain features of each layer of sample IMF components in the N layers of sample IMF components to obtain N*M sample features, where M is an integer greater than or equal to 1; The model training module is configured to train a convolutional neural network (CNN) recognition model based on N*M sample features of multiple sample DC signals and their corresponding labels to obtain a trained CNN recognition model.
12. The device according to claim 11, characterized in that The sample component selection module is configured as follows: Calculating the information entropy of each layer of sample IMF components in the multi-layer sample IMF components and the information entropy of the sample DC signal; Calculating the ratio of the information entropy of each layer of sample IMF components in the multi-layer sample IMF components to the information entropy of the sample DC signal; The N-layer sample IMF components with the largest ratio among the multi-layer sample IMF components are obtained as the N-layer sample IMF components with the largest effective information amount.
13. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to any one of claims 1 to 6.
14. A readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed by a processor, the method described in any one of claims 1 to 6 is implemented.