Fault arc detection method
Through the combination of wavelet transformation and 3D convolutional neural network model, the problems of high false alarm rate and low accuracy in fault arc diagnosis are solved, and the accurate identification of series fault arcs is achieved, and the monitoring and fault detection of power systems are improved. ability.
Patent Information
- Application Number
- CN202410991911.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
AI Technical Summary
The prior art has high false alarm rate and low accuracy in fault arc diagnosis, especially the difficulty in identifying series fault arcs, which are greatly affected by environmental noise and current waveform complexity.
Wavelet transformation is used to convert the current timing signal into a time-frequency diagram sequence, and a 3D convolutional neural network model is constructed to optimize the model by training and verification data sets to improve the accuracy of fault arc recognition.
It effectively reduces the false alarm rate of fault arc detection, improves identification accuracy and robustness, especially in complex environments, which can accurately identify series fault arcs, and enhances the monitoring and fault detection capabilities of the power system.
Smart Images

Figure CN118937915A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power systems, and in particular to a method for detecting a fault arc. Background Art
[0002] During the operation of the power system, arc faults may cause equipment damage, production interruptions, and even fires. Therefore, arc fault diagnosis technology is crucial in the power industry.
[0003] At present, the main method for diagnosing arc faults is to monitor current changes through high-precision sensors. Current transformers are usually used to convert current signals into voltage signals. The principle is to connect a sampling resistor to the output end of the current transformer and calculate the actual value of the current through Ohm's law. However, the current time series signal collected by this method is difficult to accurately identify the arc fault through the network model, especially the series arc fault. The reason is that the current change of the series arc fault is very small. Generally, the collected time series data is similar to the normal current in size and the phase is the same, so it is difficult to identify. Traditional detection methods are often affected by environmental noise and the complexity of the current waveform, resulting in a high false alarm rate and low accuracy of arc fault diagnosis. Summary of the invention
[0004] The present invention provides a method for detecting a fault arc, so as to solve the problems of high false alarm rate and low accuracy in current fault arc diagnosis.
[0005] In a first aspect, the present invention provides a method for detecting an arc fault, the method comprising:
[0006] Collect the current timing signal of the fault arc and perform preprocessing;
[0007] The current time series signal of the fault arc is transformed by wavelet to obtain a time-frequency diagram sequence;
[0008] Create training data set and validation data set based on the time-frequency graph sequence;
[0009] Build a 3D convolutional neural network model and train the 3D convolutional neural network model based on the training data set;
[0010] The time-frequency graph sequence is input into the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0011] Optionally, the current time series signal of the fault arc is subjected to wavelet transformation to obtain a time-frequency diagram sequence, including:
[0012] The current time series signal of the fault arc is segmented with a duration of 1s, and the current time series signal with a duration of 0.1s is transformed using Moret wavelet to obtain a time-frequency diagram sequence; the time-frequency diagram sequence is translated with a step size of 1 and a window width of 5.
[0013] Optionally, the current time sequence signal of the fault arc is collected and preprocessed, including:
[0014] Collect the current time series signals of fault arcs of various load types and perform noise reduction and normalization processing;
[0015] After building a 3D convolutional neural network model and training the 3D convolutional neural network model based on the training data set, it also includes:
[0016] The time-frequency graph sequence is used as the input of the 3D convolutional neural network model to obtain the current characteristic information of the fault arc;
[0017] The verification data set is input into the 3D convolutional neural network model, and the 3D convolutional neural network model is verified and its performance is evaluated based on the current characteristic information.
[0018] Optionally, create a training dataset and a validation dataset based on the time-frequency graph sequence, including:
[0019] The time-frequency graph sequence is divided into a training data set and a validation data set in a ratio of 4:1.
[0020] Optionally, build a 3D convolutional neural network model, including:
[0021] Construct a 3D convolutional neural network model consisting of convolutional layers, pooling layers, batch normalization layers, activation functions, fully connected layers, and dropout layers.
[0022] Optionally, train the 3D convolutional neural network model based on the training dataset, including:
[0023] Input the training data set to the 3D convolutional neural network model;
[0024] Extract the current characteristic information of the fault arc according to the training data set, and fuse the current characteristic information to update the parameters of the 3D convolutional neural network model;
[0025] Calculate the value of the loss function by comparing the model prediction value with the true label value;
[0026] Determine whether the difference between the model prediction value and the actual label value is less than a first preset value according to the value of the loss function;
[0027] If the difference between the model prediction value and the true label value is less than a first preset value, the learning weight of the 3D convolutional neural network model is updated.
[0028] Optionally, after determining whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function, the method further includes:
[0029] If the difference between the model prediction value and the true label value is greater than or equal to the first preset value, the current characteristic information of the fault arc is repeatedly extracted according to the training data set; and the current characteristic information is integrated to update the parameters of the 3D convolutional neural network model.
[0030] Optionally, calculate the loss function using the model predictions and true label values, including:
[0031] Calculate the value of the cross entropy loss function or the mean square error loss function based on the model prediction value and the true label value.
[0032] Optionally, update the learned weights of the 3D convolutional neural network model, including:
[0033] The learning weights of the 3D convolutional neural network model are updated using the back-propagation algorithm and the optimization algorithm.
[0034] Optionally, input a verification data set to the 3D convolutional neural network model, and verify and evaluate the performance of the 3D convolutional neural network model based on the current characteristic information, including:
[0035] The verification data set is input into the 3D convolutional neural network model, and the accuracy of the current characteristic information extraction and the fault classification quality of the fault arc of the 3D convolutional neural network model are evaluated by accuracy and root mean square error.
[0036] The technical solution of the present invention uses wavelet transform to convert the collected current time series signal of the fault arc into a time-frequency diagram sequence. The time-frequency diagram can better characterize the characteristics of the current signal changing with time compared to the original current time series signal. Therefore, the current time series signal of the fault arc and the normal current time series signal can be effectively distinguished by means of wavelet transform, and can also adapt to complex working environments and variable conditions of power systems. The introduction of the 3D convolutional neural network model further improves the performance of the classification algorithm. The 3D convolutional neural network model can extract deep current features and accurately classify using time and space relationships. The technical solution of the present invention, by combining wavelet transform and 3D convolutional neural network model, not only improves the accuracy and robustness of fault arc detection, but also greatly improves the real-time monitoring capability of the system, and effectively reduces the potential safety risks in the operation of power equipment. It overcomes the limitations and defects of traditional arc fault detection, realizes a higher perception capability of the fault arc current signal, can effectively identify the type of fault arc, especially the series fault arc, to improve the accuracy and performance of power system monitoring, fault detection and leakage location. It can be used for fault detection in actual engineering and has a wide range of uses.
[0037] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0039] Figure 1 is a flow chart of a fault arc detection method provided by an embodiment of the present invention;
[0040] Figure 2 is a flow chart of another fault arc detection method provided by an embodiment of the present invention;
[0041] Figure 3 It is a schematic diagram of converting a current time series signal into a time-frequency diagram provided by an embodiment of the present invention;
[0042] Figure 4 is a flow chart of another fault arc detection method provided by an embodiment of the present invention;
[0043] Figure 5 is a flow chart of another fault arc detection method provided by an embodiment of the present invention;
[0044] Figure 6 is a flow chart of a fault arc detection method provided by an embodiment of the present invention;
[0045] Figure 7 It is a schematic diagram of the architecture of a 3D convolutional neural network model provided by an embodiment of the present invention;
[0046] Figure 8 is a flow chart of a 3D convolutional neural network model training method provided by an embodiment of the present invention;
[0047] Fig. 9 is a flowchart of another 3D convolutional neural network model training method provided by an embodiment of the present invention;
[0048] Fig.10 is a flowchart of another 3D convolutional neural network model training method provided by an embodiment of the present invention;
[0049] Fig.11 is a flowchart of another 3D convolutional neural network model training method provided by an embodiment of the present invention;
[0050] Fig.12 It is a flow chart of another fault arc detection method provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0053] Figure 1is a flow chart of a fault arc detection method provided by an embodiment of the present invention, such as Figure 1 As shown, the method includes:
[0054] S100: collecting the current time sequence signal of the fault arc and performing preprocessing.
[0055] Specifically, the current time sequence signal of the fault arc is collected. For example, the current time sequence signal of the fault arc can be generated by an arc generating device in a laboratory, and the current time sequence signal of the fault arc can be collected by connecting a current sensor to an external resistor. The collected current time sequence signal is subjected to preprocessing operations such as denoising, normalization, and data format conversion.
[0056] S110: performing wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence.
[0057] Specifically, the collected current time series signal of the fault arc is subjected to a wavelet transform (Continuous Wavelet Transform, CWT) to obtain a time-frequency diagram sequence. Exemplarily, the current time series signal of the fault arc is subjected to a wavelet transform using Moret wavelet to obtain a time-frequency diagram sequence. Converting the current time series signal into a time-frequency diagram can better characterize the characteristics of the current signal changing over time.
[0058] S120: Creating a training data set and a validation data set according to the time-frequency graph sequence.
[0059] Specifically, the obtained time-frequency graph sequence is divided according to a certain ratio, one part of which is a training data set and the other part is a verification data set.
[0060] S130: Construct a 3D convolutional neural network model, and train the 3D convolutional neural network model based on the training data set.
[0061] Specifically, a 3D convolutional neural network model is constructed, wherein the 3D convolutional neural network model is composed of a convolutional layer, a pooling layer, a batch normalization layer, an activation function, a fully connected layer, and a random dropout layer, which is used to process and learn the feature information in the time-frequency graph sequence after wavelet transformation. After the 3D convolutional neural network model is constructed, the 3D convolutional neural network model needs to be trained. The training data set is input into the 3D convolutional neural network model, the difference between the model prediction and the true label is calculated, and the various parameters of the 3D convolutional neural network model are updated, thereby improving the performance of the 3D convolutional neural network model on the training data.
[0062] S140: Input the time-frequency graph sequence to the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0063] Specifically, the time-frequency graph sequence is input into the trained 3D convolutional neural network model. The 3D convolutional neural network model can extract characteristic information of the current of the fault arc according to the time-frequency graph sequence, thereby identifying the fault type of the fault arc and outputting it.
[0064] The technical solution of the embodiment of the present invention does not directly use the current timing signal of the fault arc, but converts the current timing signal of the fault arc into a time-frequency graph sequence through a wavelet analysis method, and inputs it into a 3D convolutional neural network model. The model encapsulates the time-frequency graph sequence into a cuboid, performs convolution operations within the cuboid, generates small batches of feature graphs, and finally fuses the features through a fully connected layer to generate a feature response graph of the corresponding category, thereby achieving the purpose of identifying the fault type.
[0065] The technical solution of the embodiment of the present invention uses wavelet transform to convert the collected current time series signal of the fault arc into a time-frequency diagram sequence. The time-frequency diagram can better characterize the characteristics of the current signal changing with time compared to the original current time series signal. Therefore, the current time series signal of the fault arc and the normal current time series signal can be effectively distinguished by means of wavelet transform, and can also adapt to complex working environments and variable conditions of power systems. The introduction of the 3D convolutional neural network model further improves the performance of the classification algorithm. The 3D convolutional neural network model can extract deep current features and accurately classify them using time and space relationships. The technical solution of the embodiment of the present invention combines wavelet transform and 3D convolutional neural network model, which not only improves the accuracy and robustness of fault arc detection, but also greatly improves the real-time monitoring capability of the system, and effectively reduces the potential safety risks in the operation of power equipment. It overcomes the limitations and defects of traditional arc fault detection, realizes a higher perception capability of the fault arc current signal, and can effectively identify the type of fault arc, especially the series fault arc, to improve the accuracy and performance of power system monitoring, fault detection and leakage location. It can be used for fault detection in practical engineering and has a wide range of uses.
[0066] Optionally, based on the above embodiments, Figure 2 is a flow chart of another arc fault detection method provided by an embodiment of the present invention. Figure 3 is a schematic diagram of converting a current time series signal into a time-frequency diagram provided by an embodiment of the present invention, such as Figure 2 As shown, the method includes:
[0067] S200: collecting the current time sequence signal of the fault arc and performing preprocessing.
[0068] S210: Divide the current time series signal of the fault arc into segments with a duration of 1 s, perform wavelet transform on the current time series signal with a duration of 0.1 s using Moret wavelet to obtain a time-frequency graph sequence; and translate the time-frequency graph sequence with a step size of 1 and a window width of 5.
[0069] Specifically, refer to Figure 3 , Figure 3 The current time series signal 1 of the fault arc with a duration of 0.5 s and the time-frequency curve obtained by wavelet analysis of the current time series signal 1 of the fault arc with a duration of 0.5 s using the Moret function are given as examples. Figure 2 The arc fault current sequence signal 1 is collected through the arc generating device in the laboratory. After processing, the arc fault current sequence signal 1 is segmented into 1s segments, and 100,000 sampling points are set for each 1s current sequence signal 1. Figure 3 The current time series signal 1 of the fault arc with a duration of 0.5s includes 50,000 sampling points. The time-frequency diagram sequence is obtained by wavelet analysis of the current time series signal 1 of each 0.1s using the Moret function. The time-frequency diagram sequence is translated with a step size of 1 and a window width of 5, and 5 time-frequency diagrams are obtained for the current time series signal 1 of the fault arc with a duration of 0.5s. Figure 2 .
[0070] S220: Create a training data set and a validation data set according to the time-frequency graph sequence.
[0071] S230: Construct a 3D convolutional neural network model, and train the 3D convolutional neural network model based on the training data set.
[0072] S240: Input the time-frequency graph sequence to the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0073] Optionally, based on the above embodiments, Figure 4 is a flow chart of another method for detecting arc faults provided by an embodiment of the present invention, such as Figure 4 As shown, the method includes:
[0074] S300: Collects the current time series signals of fault arcs of various load types and performs noise reduction and normalization processing.
[0075] Specifically, the current timing signal of the fault arc can be generated by the arc generating device in the laboratory, and the current timing signal of the fault arc can be collected by connecting the current sensor to an external resistor. The current timing signal of the fault arc generated by the arc generating device in the laboratory can be set as the current timing signal of the fault arc of various load types, ensuring the diversity of sampling. After a series of processing, the current timing signal of the fault arc of various load types outputs the fault type of the fault arc. With this setting, the fault type of the current timing signal of the fault arc of various load types can be accurately identified.
[0076] The collected current time series signal of the fault arc is subjected to denoising by the wavelet denoising method. The wavelet denoising method is based on wavelet transform, which decomposes the current time series signal of the fault arc into sub-bands of different frequencies and then removes noise by threshold processing.
[0077] Normalization processing uses the zero-mean normalization (Z-Score) method, and the specific formula is as follows:
[0078]
[0079] Wherein, x′ is the normalized current data, x is the original current time series signal, μ is the mean of the original current time series signal, and σ is the standard deviation of the original current time series signal.
[0080] S310: Perform wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence.
[0081] S320: Create a training data set and a validation data set according to the time-frequency graph sequence.
[0082] S330: Construct a 3D convolutional neural network model, and train the 3D convolutional neural network model based on the training data set.
[0083] Specifically, after training and parameter adjustment, the 3D convolutional neural network model can be applied to the time-frequency diagram sequences of fault arcs of different load types, which improves the generalization and robustness of the 3D convolutional neural network model.
[0084] S340: Using the time-frequency graph sequence as input to the 3D convolutional neural network model to obtain current characteristic information of the fault arc.
[0085] Specifically, the wavelet transform is used to process the current time series signal of the fault arc to obtain a time-frequency graph sequence. The wavelet transform can effectively decompose the current time series signal of the fault arc into different frequency components, which helps to reveal the important information hidden in the fault arc. After completing the training of the 3D convolutional neural network model, the time-frequency graph sequence is input into the 3D convolutional neural network model. Unlike the traditional 2D convolutional neural network model, the 3D convolutional neural network model can process the three-dimensional characteristics of time series data, that is, perform convolution operations in the time and space dimensions. This enables the 3D convolutional neural network model to make full use of the time series information of the current time series signal and accurately extract and analyze the key features of the current time series signal of the fault arc.
[0086] S350: Input the verification data set into the 3D convolutional neural network model, and verify and evaluate the performance of the 3D convolutional neural network model based on the current characteristic information.
[0087] Specifically, the verification data set is input into the 3D convolutional neural network model to observe whether the 3D convolutional neural network model can accurately extract the current characteristic information, so as to verify and evaluate the performance of the 3D convolutional neural network model.
[0088] S360: Input the time-frequency graph sequence to the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0089] Optionally, based on the above embodiments, Figure 5 is a flow chart of another method for detecting arc faults provided by an embodiment of the present invention, such as Figure 5 As shown, the method includes:
[0090] S400: collecting the current time sequence signal of the fault arc and performing preprocessing.
[0091] S410: Perform wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence.
[0092] S420: Divide the time-frequency graph sequence into a training data set and a validation data set in a ratio of 4:1.
[0093] Specifically, the current time series signal of the fault arc is subjected to wavelet transformation to obtain a time-frequency graph sequence, and then the time-frequency graph sequence is divided into a training data set and a verification data set in a ratio of 4:1.
[0094] S430: Construct a 3D convolutional neural network model, and train the 3D convolutional neural network model based on the training data set.
[0095] S440: Input the time-frequency graph sequence to the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0096] Optionally, based on the above embodiments, Figure 6 is a flow chart of a fault arc detection method provided by an embodiment of the present invention, such as Figure 6 As shown, the method includes:
[0097] S500: Collect the current time sequence signal of the fault arc and perform preprocessing.
[0098] S510: Perform wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence.
[0099] S520: Create a training data set and a validation data set according to the time-frequency graph sequence.
[0100] S530: Construct a 3D convolutional neural network model consisting of a convolutional layer, a pooling layer, a batch normalization layer, an activation function, a fully connected layer, and a dropout layer.
[0101] Specifically, the 3D convolutional neural network model can include multiple convolutional layers. The convolutional layers can effectively capture the spatial and temporal features of the time-frequency graph. The convolutional layers contain convolution kernels. Each convolution kernel not only considers the spatial information in the time-frequency graph, but also considers the changes in the time-frequency graph in the temporal dimension, thereby more comprehensively characterizing the dynamic characteristics of the current signal. After the convolutional layer, the pooling layer is introduced. The pooling layer helps to reduce the number of parameters of the model by reducing the spatial dimension of the feature map while retaining important feature information. The batch normalization layer is used to normalize the input of each convolutional layer, accelerate the convergence process, and enhance the robustness of the model to the input data. The activation function is applied to the output of the convolutional layer to introduce nonlinearity, so that the network model can learn more complex features and patterns. The fully connected layer is responsible for mapping the features extracted by the convolutional layer to the output layer for the final classification task. In order to reduce overfitting, the dropout layer can also be introduced. The dropout layer randomly sets the output of some neurons to zero during the training process to force the network to learn more generalized feature representations.
[0102] Figure 7 is a schematic diagram of the architecture of a 3D convolutional neural network model provided by an embodiment of the present invention, such as Figure 7 As shown, Figure 7The architecture of a 3D convolutional neural network model consisting of a hard-wired layer H1, a convolutional layer C2, a pooling layer S3, a convolutional layer C4, a pooling layer S5, a convolutional layer C6, and a fully connected layer F7 is given as an example. Input: 7@60*40 represents the input of 7 consecutive frames of 60*40 time-frequency graphs, H1: 33@60*40 represents that the hard-wired layer H1 contains 33 60*40 time-frequency graphs, 7*7*3 represents the convolution kernel size, C2: 23*2@54*34 represents that the convolution layer C2 has 2 7*7*3 convolution kernels, which are used to convolve the time-frequency graph of the hard-wired layer H1 to obtain 46 feature maps with a size of 54*34. 2*2 represents the window size of the pooling layer S3, S3: 23*2@27*17 means that the pooling layer S3 uses a 2*2 window for pooling, and the feature The number of images remains unchanged, and the size of each feature map is 27*17. 7*6*3 represents the convolution kernel size. C4: 13*6@21*12 means that the convolution layer C4 has 3 convolution kernels of size 7*6*3, the number of feature maps is 13*6, and the size of each image is 21*12. 3*3 represents the window size of the pooling layer S5. S5: 13*6@7*4 means that the window size of the pooling layer S5 is 3*3, the number of feature maps remains unchanged, but the size becomes 7*4. 7*4 represents the convolution kernel size. C6: 128@1*1 means that the convolution kernel size of the convolution layer C6 is 7*4, the feature map size is 1*1, and the number of feature maps is 128.
[0103] S540: Input the time-frequency graph sequence to the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0104] Optionally, based on the above embodiments, Figure 8 is a flow chart of a 3D convolutional neural network model training method provided by an embodiment of the present invention, such as Figure 8 As shown, the method includes:
[0105] S600: Input the training data set to the 3D convolutional neural network model.
[0106] Specifically, after constructing the 3D convolutional neural network model, the 3D convolutional neural network model is trained. The training data set may be preprocessed first, and exemplary preprocessing operations such as noise reduction may be included. The preprocessed training data set is input into the 3D convolutional neural network model.
[0107] S610: Extracting current characteristic information of the fault arc according to the training data set, and fusing the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0108] Specifically, the 3D convolutional neural network model can perform feature extraction on the training data set, extract the current characteristic information of the fault arc according to the training data set, and change the parameter setting of the 3D convolutional neural network model accordingly according to the current characteristic information. After changing the parameter setting of the 3D convolutional neural network model, a model prediction value can be generated.
[0109] S620: Calculate the value of the loss function using the model prediction value and the true label value.
[0110] Specifically, a suitable loss function is selected. For example, a cross entropy loss function can be selected. The loss function is used for classification tasks. The loss function is calculated according to the model prediction value and the true label value to obtain the value of the loss function.
[0111] S630: Determine whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function.
[0112] Specifically, it is determined according to the value of the loss function whether the difference between the model prediction value and the true label value is less than a first preset value, where the first preset value is preset according to actual conditions. If the difference between the model prediction value and the true label value is less than the first preset value, S641 is executed.
[0113] S641: Update the learning weights of the 3D convolutional neural network model.
[0114] Specifically, the difference between the model prediction value and the true label value is less than the first preset value, the 3D convolutional neural network model obtains gradient information from the loss function, updates the learning weights and biases of the 3D convolutional neural network model to minimize the value of the loss function, and improves the performance of the 3D convolutional neural network model on the training data set.
[0115] Optionally, based on the above embodiments, Fig. 9 is a flowchart of another 3D convolutional neural network model training method provided by an embodiment of the present invention, such as Fig. 9 As shown, the method includes:
[0116] S700: Input the training data set to the 3D convolutional neural network model.
[0117] S710: Extracting current characteristic information of the fault arc according to the training data set, and fusing the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0118] S720: Calculate the value of the loss function using the model prediction value and the true label value.
[0119] S730: Determine whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function.
[0120] S741: Update the learning weights of the 3D convolutional neural network model.
[0121] S742: Repeatedly extract the current characteristic information of the fault arc according to the training data set; and fuse the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0122] Specifically, if the difference between the model prediction value and the true label value is greater than or equal to the first preset value, the current characteristic information of the fault arc is repeatedly extracted through the training data set; and the current characteristic information is integrated to update the parameters of the 3D convolutional neural network model. The parameters of the 3D convolutional neural network model are optimized by iterating the entire data set.
[0123] After a large number of iterative training, the 3D convolutional neural network model can identify the fault category based on the time-frequency diagram sequence of the fault arc. The wavelet transform and 3D convolutional neural network model can be used as a whole recognition system. It can achieve the purpose of identifying the fault type by collecting the current timing signal of the fault arc, thus expanding the scope of use on the engineering site.
[0124] Optionally, based on the above embodiments, Fig.10 is a flowchart of another 3D convolutional neural network model training method provided by an embodiment of the present invention, such as Fig.10 As shown, the method includes:
[0125] S800: Input the training data set to the 3D convolutional neural network model.
[0126] S810: Extracting current characteristic information of the fault arc according to the training data set, and fusing the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0127] S820: Calculate the value of the cross entropy loss function or the mean square error loss function according to the model prediction value and the true label value.
[0128] Specifically, a cross entropy loss function or a mean square error loss function is selected as the loss function, and the loss function is used for classification tasks. The cross entropy loss function or the mean square error loss function is calculated according to the model prediction value and the true label value to obtain the value of the cross entropy loss function or the mean square error loss function.
[0129] S830: Determine whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function.
[0130] S841: Update the learning weights of the 3D convolutional neural network model.
[0131] S842: Repeatedly extract the current characteristic information of the fault arc according to the training data set; and fuse the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0132] Optionally, based on the above embodiments, Fig.11 is a flowchart of another 3D convolutional neural network model training method provided by an embodiment of the present invention, such as Fig.11 As shown, the method includes:
[0133] S900: Input the training data set to the 3D convolutional neural network model.
[0134] S910: Extracting current characteristic information of the fault arc according to the training data set, and fusing the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0135] S920: Calculate the value of the loss function using the model prediction value and the true label value.
[0136] S930: Determine whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function.
[0137] S941: Update the learning weights of the 3D convolutional neural network model using back-propagation and optimization algorithms.
[0138] Specifically, the difference between the model prediction value and the true label value is less than the first preset value, the 3D convolutional neural network model obtains gradient information from the loss function through the back propagation algorithm, and then uses the optimization algorithm to update the learning weights and biases of the 3D convolutional neural network model to minimize the value of the loss function and improve the performance of the 3D convolutional neural network model on the training data set. Exemplarily, the optimization algorithm can be an adaptive moment estimation (Adaptive Moment Estimation, Adam) algorithm.
[0139] S942: Repeatedly extract the current characteristic information of the fault arc according to the training data set; and fuse the current characteristic information to update the parameters of the 3D convolutional neural network model.
[0140] Optionally, based on the above embodiments, Fig.12 is a flow chart of another method for detecting arc faults provided by an embodiment of the present invention, such as Fig.12 As shown, the method includes:
[0141] S1000: Collects the current time series signals of fault arcs of various load types and performs noise reduction and normalization processing.
[0142] S1010: Perform wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence.
[0143] S1020: Create a training data set and a validation data set according to the time-frequency graph sequence.
[0144] S1030: Construct a 3D convolutional neural network model, and train the 3D convolutional neural network model based on the training data set.
[0145] S1040: Using the time-frequency graph sequence as input to the 3D convolutional neural network model to obtain current characteristic information of the fault arc.
[0146] S1050: Input the validation data set into the 3D convolutional neural network model, and evaluate the accuracy of the current characteristic information extraction of the 3D convolutional neural network model and the fault classification quality of the fault arc through the accuracy and root mean square error. The accuracy and root mean square error are used to quantify the 3D convolutional neural network model's ability to extract the time-frequency diagram feature information of the fault arc and the accuracy of identifying the fault type.
[0147] Specifically, the verification data set is input into the 3D convolutional neural network model, and the accuracy and root mean square error are used to observe whether the 3D convolutional neural network model can accurately extract the current characteristic information, so as to verify and evaluate the performance of the 3D convolutional neural network model.
[0148] The calculation formula of accuracy is as follows:
[0149]
[0150] Among them, Acc represents the accuracy rate, TP represents the number of samples where both the model prediction value and the true label value are positive; TN represents the number of samples where both the model prediction value and the true label value are negative; FP represents the number of samples where the model prediction value is positive and the true label value is negative; FN represents the number of samples where the model prediction value is negative and the true label value is positive.
[0151] The formula for calculating the root mean square error is as follows:
[0152]
[0153] Among them, the root mean square error RMSE is a common indicator to measure the difference between the model prediction value and the true label value. Where N represents the number of time steps of the time series signal, X i represents the true label value generated at the current moment i; Y i Represents the model prediction value at the current time i.
[0154] S1060: Input the time-frequency graph sequence to the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
[0155] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.
[0156] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for detecting a fault arc, characterized in that: include: Collect the current timing signal of the fault arc and perform preprocessing; Performing wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence; Creating a training data set and a validation data set according to the time-frequency graph sequence; Constructing a 3D convolutional neural network model, and training the 3D convolutional neural network model based on the training data set; The time-frequency graph sequence is input into the 3D convolutional neural network model, and the 3D convolutional neural network model outputs the fault type of the fault arc.
2. The method for detecting arc fault according to claim 1, characterized in that: The step of performing wavelet transformation on the current time series signal of the fault arc to obtain a time-frequency diagram sequence comprises: The current time series signal of the fault arc is segmented with a duration of 1 s, and the current time series signal with a duration of 0.1 s is subjected to wavelet transformation using Moret wavelet to obtain the time-frequency diagram sequence; the time-frequency diagram sequence is translated with a step size of 1 and a window width of 5.
3. The method for detecting arc fault according to claim 1, characterized in that: The collecting of the current time sequence signal of the fault arc and preprocessing includes: Collecting the current time sequence signals of the fault arc of various load types and performing noise reduction and normalization processing; After constructing the 3D convolutional neural network model and training the 3D convolutional neural network model based on the training data set, the method further includes: Using the time-frequency graph sequence as input of the 3D convolutional neural network model to obtain current characteristic information of the fault arc; The verification data set is input into the 3D convolutional neural network model, and the 3D convolutional neural network model is verified and performance evaluated based on the current characteristic information.
4. The method for detecting arc fault according to claim 1, characterized in that: The step of creating a training data set and a verification data set according to the time-frequency graph sequence comprises: The time-frequency graph sequence is divided into the training data set and the verification data set in a ratio of 4:
1.
5. The arc fault detection method according to claim 1, characterized in that: The 3D convolutional neural network model is constructed, comprising: Construct a 3D convolutional neural network model consisting of convolutional layers, pooling layers, batch normalization layers, activation functions, fully connected layers, and dropout layers.
6. The arc fault detection method according to claim 1, characterized in that: The training of the 3D convolutional neural network model based on the training data set includes: Inputting the training data set into the 3D convolutional neural network model; Extracting current characteristic information of the fault arc according to the training data set, and fusing the current characteristic information to update the parameters of the 3D convolutional neural network model; Calculate the value of the loss function by comparing the model prediction value with the true label value; Determine whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function; If the difference between the model prediction value and the true label value is less than the first preset value, the learning weight of the 3D convolutional neural network model is updated.
7. The arc fault detection method according to claim 6, characterized in that: After judging whether the difference between the model prediction value and the true label value is less than a first preset value according to the value of the loss function, the method further includes: If the difference between the model prediction value and the true label value is greater than or equal to the first preset value, the current characteristic information of the fault arc is repeatedly extracted according to the training data set; and the current characteristic information is integrated to update the parameters of the 3D convolutional neural network model.
8. The arc fault detection method according to claim 6, characterized in that: The calculating the value of the loss function by using the model prediction value and the true label value includes: The value of the cross entropy loss function or the mean square error loss function is calculated according to the model prediction value and the true label value.
9. The arc fault detection method according to claim 6, characterized in that: The updating of the learning weights of the 3D convolutional neural network model comprises: The learning weights of the 3D convolutional neural network model are updated using a back-propagation algorithm and an optimization algorithm.
10. The arc fault detection method according to claim 3, characterized in that: The inputting the verification data set into the 3D convolutional neural network model, and verifying and evaluating the performance of the 3D convolutional neural network model according to the current characteristic information, includes: The verification data set is input into the 3D convolutional neural network model, and the accuracy of the current characteristic information extraction of the 3D convolutional neural network model and the fault classification quality of the fault arc are evaluated by accuracy and root mean square error.
Citation Information
Cited By
Fault diagnosis method and system for floor radiation heating system, medium and equipment
CN119538104A
Long-distance arc detection method
CN121904515A
