Step-by-step fault arc detection method and device, electronic equipment, and storage medium

Abstract

The present invention relates to the field of arc detection technology, and discloses a step-by-step fault arc detection method and device, electronic equipment, and storage medium, including: extracting time domain characteristic parameters from a current signal; judging the change value between the consecutive cycles before and after the time domain characteristic parameters; if the change value is less than a set threshold, it is considered that the detection result of the current cycle is the same as that of the previous cycle; if the change value is greater than the set threshold, the current signal is detected by a trained fault arc detection convolutional neural network to obtain a first detection result, the current signal is detected by discrete wavelet transform and the detection result is compared with an adaptive threshold to obtain a second detection result; the first detection result is calculated to obtain a first weight, the second detection result is calculated by an adaptive threshold to obtain a second weight, and then whether a fault arc occurs is judged by weight arbitration, statistics, and comparison. It has the characteristics of high fault arc detection accuracy and simple calculation process.

Description

Technical Field

[0001] The present application relates to the field of arc detection technology, and in particular to a step-by-step fault arc detection method and device, electronic equipment, and storage medium. Background Art

[0002] Fire accidents caused by electrical fires are generally more serious, and reducing the occurrence of electrical fires is an urgent problem. Line causes can be roughly divided into the following two categories based on the causes:

[0003] (1) Electrical fires caused by abnormal currents. For example, if a circuit is short-circuited or overloaded, the current on the circuit surges in a short period of time, causing a large amount of heat to be generated, melting the insulation layer and igniting surrounding flammable materials, leading to a fire.

[0004] (2) Electrical fires caused by fault arcs. Fault arcs are essentially electrical breakdowns caused by abnormal voltage, which generates a lot of heat and causes fires. Poor contact between wires, wear of wire insulation, and other reasons may lead to electrical fires.

[0005] In the first case, it is relatively easy to detect a surge in current, and various fuse breakers and air switches have been developed to prevent casualties and fires. However, electrical fires caused by fault arcs are difficult to detect due to their diversity, randomness, and complexity.

[0006] The type of load determines the waveform of the current in the branch. The normal current waveform and fault arc current waveform of different loads are very different. The flat shoulder feature of the fault arc current waveform of some loads is very obvious, and the waveform of the fault arc of some loads is mainly high-frequency pulses.

[0007] Although there are many types of fault arc detection algorithms, they all have certain defects. Although the method of detecting the physical phenomenon when the fault arc is generated has extremely high accuracy and real-time performance, it is greatly limited by the location of the fault arc; the detection method of extracting signal features in the time domain for the voltage or current waveform when the fault arc occurs is simple to implement, but can only reflect part of the characteristics of the signal waveform and has low accuracy; the accuracy of the detection method of converting the current signal to the frequency domain through short-time Fourier transform, wavelet transform and other time-frequency domain methods, and detecting the high-frequency noise generated by the fault arc in the frequency domain is highly correlated with the setting of the threshold signal; and the detection method of directly training the neural network on the fault arc current waveform has extremely high accuracy for fully trained load detection, but for untrained loads, the detection accuracy is difficult to guarantee. Secondly, the false alarm situation in the neural network detection method is relatively serious. Summary of the invention

[0008] The purpose of the embodiments of the present application is to provide a step-by-step fault arc detection method and device, electronic equipment, and storage medium to solve the technical problems of low fault arc detection accuracy and complex calculation process existing in the related art.

[0009] According to a first aspect of an embodiment of the present application, a step-by-step arc fault detection method is provided, characterized in that it includes:

[0010] Acquiring a current signal to be detected;

[0011] Extracting time domain characteristic parameters from the current signal;

[0012] Determine the change value between the preceding and following consecutive cycles according to the time domain characteristic parameter;

[0013] If the change value is less than the set threshold, it is considered that the detection result of the current cycle is the same as that of the previous cycle;

[0014] If the change value is greater than the set threshold, the current signal is detected by a trained fault arc detection convolutional neural network to obtain a first detection result, and the current signal is detected by discrete wavelet transform and the detection result is compared with an adaptive threshold to obtain a second detection result;

[0015] The first detection result is calculated by a softmax function to obtain a first weight, the second detection result is calculated by an adaptive threshold to obtain a second weight, and an accurate detection result is obtained through weight arbitration;

[0016] According to the precise detection result, the number of fault cycles occurring within a unit time is counted;

[0017] Compare the quantity with the national standard requirements to determine whether a fault arc occurs.

[0018] Furthermore, the current signal is an alternating current signal connected in series in the load circuit, which should be 50 Hz, 220V alternating current according to domestic standards.

[0019] Furthermore, the time domain characteristic parameter is selected from one or more of peak-to-peak value, rectified mean value, effective value and variance.

[0020] Furthermore, the construction method of the trained arc fault detection convolutional neural network is as follows:

[0021] Acquire a current signal data set with an arc and a current signal data set without an arc;

[0022] The convolutional neural network is trained by using the current signal data set with arc and the current signal data set without arc to obtain a trained fault arc detection convolutional neural network.

[0023] Furthermore, the arc fault detection convolutional neural network includes:

[0024] The first convolution layer is used to perform convolution calculation on the input current signal and extract the input current signal characteristics;

[0025] The first maximum pooling layer is used to perform maximum pooling on the calculation results of the first convolutional layer and extract the signal features of the first convolutional layer;

[0026] The second convolutional layer is used to perform convolution calculation on the result of the first maximum pooling layer to extract the features of the result of the first maximum pooling layer;

[0027] The second maximum pooling layer is used to perform maximum pooling on the calculation results of the second convolutional layer and extract the signal features of the second convolutional layer;

[0028] The third convolutional layer is used to perform convolution calculation on the result of the second maximum pooling layer and extract the features of the result of the second maximum pooling layer;

[0029] The third maximum pooling layer is used to perform maximum pooling on the calculation results of the third convolutional layer and extract the signal features of the third convolutional layer;

[0030] The tiling layer is used to combine and sort the calculation results of the third maximum pooling layer;

[0031] The fully connected layer is used to further calculate the results of the tiled layer and output the calculation results of the corresponding labels as the first detection results.

[0032] Further, the current signal is detected by discrete wavelet transform and the detection result is compared with the adaptive threshold to obtain a second detection result, including:

[0033] Using Db4 wavelet as mother wavelet, discrete wavelet transform is performed on the input fault arc current signal;

[0034] The maximum value max of the statistical discrete wavelet transform result;

[0035] Perform adaptive threshold calculation on discrete wavelet transform results;

[0036] The second detection result is obtained by comparing the maximum value of the discrete wavelet transform detection result with the calculated adaptive threshold. If the maximum value of the discrete wavelet transform detection result is greater than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc; if the maximum value of the discrete wavelet transform detection result is less than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc.

[0037] Further, the first detection result is calculated by a softmax function to obtain a first weight, the second detection result is calculated by an adaptive threshold to obtain a second weight, and an accurate detection result is obtained by weight arbitration, including:

[0038] Calculating the first detection result of the fault arc convolutional neural network detection by a softmax function to obtain a first weight value of the convolutional neural network detection result;

[0039] Calculate the second detection result of discrete wavelet transform detection by using an adaptive threshold value, and obtain a second weight value of the discrete wavelet transform detection result;

[0040] The first detection result and the first weight value are respectively combined with the second detection result and the second weight value, and an accurate detection result of the fault arc is obtained through weight arbitration.

[0041] According to a second aspect of an embodiment of the present invention, there is provided a step-by-step fault arc detection device, comprising:

[0042] An acquisition module, used for acquiring a current signal to be detected;

[0043] An extraction module, used for extracting time domain characteristic parameters from the current signal;

[0044] A judgment module, used for judging the change value between the preceding and following consecutive cycles according to the time domain characteristic parameter;

[0045] A first judgment submodule, configured to consider that the detection result of the current cycle is the same as that of the previous cycle if the change value is less than a set threshold;

[0046] A second judgment submodule is used for, if the change value is greater than a set threshold, detecting the current signal through a trained fault arc detection convolutional neural network to obtain a first detection result, detecting the current signal through discrete wavelet transform and comparing the detection result with an adaptive threshold to obtain a second detection result;

[0047] A calculation module, configured to calculate the first detection result by using a softmax function to obtain a first weight, calculate the second detection result by using an adaptive threshold to obtain a second weight, and obtain an accurate detection result by weight arbitration;

[0048] A statistical module, used for counting the number of fault cycles occurring within a unit time according to the precise detection result;

[0049] The comparison and judgment module is used to compare the quantity with the national standard requirements to determine whether a fault arc occurs.

[0050] According to a third aspect provided by an embodiment of the present invention, there is provided an electronic device, including:

[0051] one or more processors;

[0052] A memory for storing one or more programs;

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0054] According to a fourth aspect provided by an embodiment of the present invention, there is provided a computer-readable storage medium having computer instructions stored thereon, wherein the instructions, when executed by a processor, implement the steps of the method described in the first aspect.

[0055] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0056] It can be seen from the above technical solutions that the present application adopts a technical means combining convolutional neural network detection and adaptive threshold discrete wavelet transform detection, which overcomes the false alarm situation in convolutional neural network detection, thereby achieving the technical effect of further improving the accuracy. The pre-detection and precise detection step-by-step structure is adopted, so as to overcome the technical problem that the simplification and accuracy of the fault arc detection algorithm are difficult to coexist, thereby achieving the technical effect of simplifying the calculation process of fault arc detection while ensuring high accuracy.

[0057] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0059] Figure 1 The figure is a flow chart showing a step-by-step arc fault detection method according to an exemplary embodiment.

[0060] Figure 2 is a diagram showing normal and fault arc current waveforms of different loads according to an exemplary embodiment.

[0061] Figure 3 is a waveform diagram of mother wavelet Db4 according to an exemplary embodiment.

[0062] Figure 4 It is a schematic structural diagram of a step-by-step fault arc detection device according to an exemplary embodiment. DETAILED DESCRIPTION

[0063] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0064] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0065] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0066] Figure 1 is a flow chart of a step-by-step fault arc detection method according to an exemplary embodiment. Figure 1 As shown, the method is applied in a terminal and may include the following steps:

[0067] Step S11, obtaining a current signal to be detected;

[0068] Step S12, extracting time domain characteristic parameters from the current signal;

[0069] Step S13, determining the change value between the consecutive cycles according to the time domain characteristic parameter;

[0070] Step S14, if the change value is less than the set threshold, it is considered that the detection result of the current cycle is the same as that of the previous cycle;

[0071] Step S15, if the change value is greater than the set threshold, the current signal is detected by a trained fault arc detection convolutional neural network to obtain a first detection result, and the current signal is detected by discrete wavelet transform and the detection result is compared with an adaptive threshold to obtain a second detection result;

[0072] Step S16, calculating the first detection result by a softmax function to obtain a first weight, calculating the second detection result by an adaptive threshold to obtain a second weight, and obtaining an accurate detection result by weight arbitration;

[0073] Step S17, counting the number of fault cycles occurring within a unit time according to the accurate detection result;

[0074] Step S18, comparing the quantity with the national standard requirements to determine whether a fault arc occurs.

[0075] It can be seen from the above technical solutions that the present application adopts a technical means combining convolutional neural network detection and adaptive threshold discrete wavelet transform detection, which overcomes the false alarm situation in convolutional neural network detection, thereby achieving the technical effect of further improving the accuracy. The use of a step-by-step structure for pre-detection and precise detection overcomes the technical problem that the simplification of the fault arc detection algorithm and the difficulty in coexisting with the accuracy, thereby achieving the technical effect of simplifying the calculation process of the fault arc detection while ensuring high accuracy. The real-time step-by-step fault arc detection method provided by the embodiment of the present invention is suitable for real-time detection of series fault arcs of various types of loads in the home.

[0076] In the specific implementation of step S11, a current signal to be detected is obtained;

[0077] The step-by-step fault arc detection method of the present invention performs the first step of acquiring the current signal to be detected when performing fault arc detection. When acquiring the current signal, the current signal in any load circuit can be acquired as the detection current signal through an external circuit. Furthermore, the sampling ADC satisfies the sampling rate of 1M / s and the accuracy is high enough. The high sampling rate and high accuracy can make the signal features contained in the acquired current signal to be detected more and more obvious.

[0078] In a specific implementation of step S12, time domain characteristic parameters are extracted from the current signal;

[0079] After the ADC samples the current signal, it intercepts a cycle length of the current signal for detection. After obtaining the detection current signal, the current signal is first pre-detected. Pre-detection is to extract the time domain characteristic parameters of the current signal. Figure 2As shown, when the load is working normally, the AC current signal presents complete periodicity, and the time domain characteristic parameters are selected from one or more of the peak-to-peak value, rectified average value, effective value and variance. However, when a fault arc occurs, these four parameters will change dramatically due to the presence of flat shoulders and high-frequency noise pulses in the fault arc. By determining whether the time domain characteristic parameters between adjacent cycles are too different, it can be preliminarily determined whether the current signal has changed. If the current signal has not changed, it indicates that there must be no fault arc. If the current signal has changed, further precise detection is required. In fault arc detection, the vast majority of current signals are normal, and only part of the signals are fault arc signals. Using time domain characteristic parameters that are sensitive to changes in current signals as a pre-detection method to distinguish suspected fault arc current signals from normal current signals can simplify the calculation process of the detection method.

[0080] In the specific implementation of step S13, the change value between the previous and next consecutive cycles is determined according to the time domain characteristic parameter; and in the specific implementation of step S14, if the change value is less than the set threshold, it is considered that the detection result of the current cycle is the same as that of the previous cycle;

[0081] One of the time domain characteristic parameters of pre-detection is the peak-to-peak value of the current signal. Taking the fan current signal as an example, when the fan is working normally without a fault arc, the peak-to-peak value of the fan current signal is about -0.2A-0.2A, and has complete periodicity. However, when a fault arc occurs, the peak-to-peak value of the fan current signal surges to about -4A-4A. It can be seen that it is feasible to use the peak-to-peak value to pre-detect the fault arc.

[0082] One of the time domain characteristic parameters of pre-detection is the average value of the current signal. Taking the electric heating platform current signal as an example, when the electric heating platform works normally without a fault arc, the peak of the electric heating platform current signal presents a relatively complete sine wave shape, and has complete periodicity, and the average value remains unchanged. However, when a fault arc occurs, the electric heating platform current signal will have a period of time when the current is close to 0, which is called the flat shoulder stage, causing its average value to drop. It can be seen that it is feasible to use the average value to pre-detect the fault arc.

[0083] One of the time domain characteristic parameters of pre-detection is the root mean square value of the current signal. Taking the electric heating platform current signal as an example, when the electric heating platform works normally without a fault arc, the peak of the electric heating platform current signal presents a relatively complete sine wave shape, and has complete periodicity, and the root mean square value remains unchanged. However, when a fault arc occurs, the electric heating platform current signal will have a period of time when the current is close to 0, which is called the flat shoulder stage, causing its root mean square value to decrease. It can be seen that it is feasible to use the root mean square value to pre-detect the fault arc.

[0084] One of the time domain characteristic parameters of pre-detection is the variance of the current signal. Taking the electric heating station current signal as an example, when the electric heating station works normally without a fault arc, the peak of the electric heating station current signal presents a relatively complete sine wave shape, and has complete periodicity, and the variance value remains unchanged. However, when a fault arc occurs, the electric heating station current signal will have a period of time when the current is close to 0, which is called the flat shoulder stage, causing its variance value to decrease. It can be seen that it is feasible to use the variance value to pre-detect the fault arc.

[0085] If the peak-to-peak value, average value, root mean square, variance, etc. obtained from the pre-test results remain basically unchanged, and the error range is within 30%, it is considered that the characteristic values ​​of the two cycles before and after have not changed abnormally, and there is no need to conduct further precise detection of the current waveform. The detection result of the current waveform of this cycle remains unchanged from the detection result of the current waveform of the previous cycle. If the result of the previous cycle is normal, the detection result of this cycle is also considered normal; if the detection result of the previous cycle is that there is an arc, the detection result of this cycle is also considered to be that there is an arc; but the probability of the second situation is very small, because when an arc is generated, its physical characteristics cause the current waveform to have a certain numerical difference even if there are two adjacent arcs, and the detection results of the previous and next cycles are also likely to be abnormal.

[0086] If the peak-to-peak value, average value, root mean square value, variance, etc. obtained from the pre-detection results change dramatically and the error range is greater than 30%, it is considered that the characteristic values ​​of the two cycles before and after have changed abnormally, but the cause of the abnormal change is unknown. It may be caused by an arc, or it may be other normal conditions, such as the appliance switching working mode, or a new appliance being added to the load, etc. It is necessary to further accurately detect the current waveform and use the result of the accurate detection as the fault arc detection result of the current signal.

[0087] In the specific implementation of step S15, if the change value is greater than the set threshold, the current signal is detected by a trained fault arc detection convolutional neural network to obtain a first detection result, and the current signal is detected by discrete wavelet transform and the detection result is compared with an adaptive threshold to obtain a second detection result;

[0088] When entering precise detection, convolutional neural network detection and adaptive threshold discrete wavelet transform detection will be performed simultaneously.

[0089] According to the national standard, the fault arc count is in units of half-sine wave cycles. Under the premise that the frequency of civil AC power in my country is 50Hz and the sampling frequency of the front ADC is 1M, the input data set of the neural network in this paper is half-sine wave cycle data in units of 10,000 points. The collected data is cropped and labeled, and finally the results are shown in Table 1 below.

[0090] In this embodiment, the method for constructing the trained arc fault detection convolutional neural network is as follows:

[0091] (1) obtaining a current signal data set with an arc and a current signal data set without an arc;

[0092] (2) A convolutional neural network is trained using the current signal data set with arc and the current signal data set without arc to obtain a trained fault arc detection convolutional neural network.

[0093] Table 1 Convolutional neural network dataset

[0094]

[0095]

[0096] There are a total of 11,921 data items. After random shuffling, about 80% are divided into the data set (9,621 items), and about 20% are divided into the test set (2,300 items).

[0097] In the step-by-step fault arc detection method, the fault arc current data is a matrix of 1x10000. Based on the LeNet model and the VGG-16 model, the model shown in Table 2 below is modified and designed.

[0098] In this embodiment, the arc fault detection convolutional neural network includes:

[0099] The first convolution layer is used to perform convolution calculation on the input current signal and extract the input current signal characteristics;

[0100] The first maximum pooling layer is used to perform maximum pooling on the calculation results of the first convolutional layer and extract the signal features of the first convolutional layer;

[0101] The second convolutional layer is used to perform convolution calculation on the result of the first maximum pooling layer to extract the features of the result of the first maximum pooling layer;

[0102] The second maximum pooling layer is used to perform maximum pooling on the calculation results of the second convolutional layer and extract the signal features of the second convolutional layer;

[0103] The third convolutional layer is used to perform convolution calculation on the result of the second maximum pooling layer and extract the features of the result of the second maximum pooling layer;

[0104] The third maximum pooling layer is used to perform maximum pooling on the calculation results of the third convolutional layer and extract the signal features of the third convolutional layer;

[0105] The tiling layer is used to combine and sort the calculation results of the third maximum pooling layer;

[0106] The fully connected layer is used to further calculate the results of the tiled layer and output the calculation results of the corresponding labels as the first detection results.

[0107] Table 2 Convolutional neural network model

[0108]

[0109]

[0110] The model has the following characteristics: 1. The model contains 3 convolution layers activated by Relu function. The convolution kernel size used in the convolution layer is 1x5, the step size is 1, and the padding mode of automatic zero filling (padding mode = SAME) is used to ensure that the output layer model has the same structure as the input layer model. 2. The second convolution layer is converted from one channel to four channels, and the third convolution layer is converted from four channels to eight channels. As mentioned above, convolution is the process of extracting local features. More channels can extract more local features, but at the same time, in order to reduce the model size and the number of parameters, the model depth and the number of channels cannot be increased indefinitely. The number of channels and the model depth of this model are the best structures that compromise between accuracy and model size after multiple tests. 3. Like LeNet, each convolution layer is followed by a maximum pooling layer. In this model, since the sampling rate is much greater than the rate of change of signal features, the pooling step size used is larger, which can further reduce the size and number of parameters of the convolutional neural network model. 4. The 8th layer is a fully connected layer, and the activation function used is softmax. The function of this layer is to correspond to each output result according to the result of the previous tiled layer. In this model, there are 9 output results in total, corresponding to 9 channels. After the fully connected layer, each channel will have a result indicating the closeness between the input and the result. The larger the data value, the closer it is. The training of the convolutional neural network model is carried out on the TensorFlow platform. On the TensorFlow platform, the convolutional neural network model is built according to Table 2. During training, the optimizer used is "adam", and the training is carried out with accuracy as the evaluation standard. The size of a single batch is 50 groups of data, a total of 1000 training times, and the training results are saved. The training results show that the accuracy of the validation set can reach 99.17%.

[0111] In the precise detection step, the current signal to be detected is input into the fault arc detection convolutional neural network trained by the above method, and the convolutional neural network is used to calculate the result to obtain the first detection result.

[0112] In this embodiment, the current signal is detected by discrete wavelet transform and the detection result is compared with the adaptive threshold to obtain a second detection result, including:

[0113] (1) Using Db4 wavelet as the mother wavelet, the input fault arc current signal is subjected to discrete wavelet transform;

[0114] (2) Calculate the maximum value max of the discrete wavelet transform result;

[0115] (3) Adaptive threshold calculation is performed on discrete wavelet transform results;

[0116] (4) Compare the maximum value of the discrete wavelet transform detection result with the calculated adaptive threshold to obtain a second detection result. If the maximum value of the discrete wavelet transform detection result is greater than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc; if the maximum value of the discrete wavelet transform detection result is less than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc.

[0117] In the discrete wavelet transform detection method, one of the most important steps is the mother wavelet selection. For wavelet transform, different mother wavelets will produce completely different results for the detection of the same signal. Mother wavelets usually have properties such as orthogonality, compact support, symmetry and vanishing moment. When selecting a mother wavelet, the properties of the mother wavelet should be fully considered. However, there are often more than one mother wavelet with the same properties. In order to solve this problem, the similarity between the signal and the mother wavelet is considered when selecting the mother wavelet. According to the principle of similarity between the signal and the mother wavelet, as follows Figure 3 As shown in the Db4 mother wavelet waveform, when a fault arc occurs, the load current waveform has not only high-frequency pulses, but also often has the characteristics of a flat shoulder. The shape of the flat shoulder is similar to that of the Db4 mother wavelet, so Db4 is selected as the mother wavelet.

[0118] In fault arc detection, the threshold setting of discrete wavelet transform is also one of the keys to determine the discrete wavelet transform detection method. By analyzing the detection results of fixed threshold discrete wavelet transform, it is found that in the fault arc detection method of fixed threshold discrete wavelet transform, in order to avoid false alarms, the detection threshold of wavelet transform is set relatively high, resulting in many cases where the fault arc is not severe, and the discrete wavelet transform results cannot reach the threshold, so it is missed. This paper proposes an adaptive threshold algorithm in wavelet transform. First, a basic threshold max / 2 is determined according to the amplitude of the discrete wavelet transform detection result, and then the threshold is adjusted according to the discrete degree of the signal. The lower the discrete degree, the better the signal regularity, the less likely it is to cause a fault arc, and the larger the threshold should be; the higher the discrete degree, the worse the signal regularity, the more likely it is to cause a fault arc, and the smaller the threshold should be. Therefore, the threshold is inversely proportional to the average deviation. The final result is expressed by the following formula. Among them, max represents the maximum value of the discrete wavelet transform detection result, abs_dev represents the discrete degree, and abs_avg represents the average value. K represents a custom parameter, which is set to 5 here, x i Represents the input value.

[0119]

[0120]

[0121]

[0122] The second detection result is obtained by comparing the maximum value of the discrete wavelet transform detection result with the calculated adaptive threshold. If the maximum value of the discrete wavelet transform detection result is greater than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc; if the maximum value of the discrete wavelet transform detection result is less than the adaptive threshold, the second detection result is judged as the occurrence of no fault arc.

[0123] In the specific implementation of step S16, the first detection result is calculated by a softmax function to obtain a first weight, the second detection result is calculated by an adaptive threshold to obtain a second weight, and an accurate detection result is obtained through weight arbitration;

[0124] The adaptive threshold discrete wavelet transform detection method has a better effect on fault arc detection with high-frequency pulses, while the convolutional neural network detection method has a better performance in detecting flat shoulders. In order to comprehensively consider the detection results of convolutional neural network and adaptive threshold discrete wavelet transform, an arbitration method is proposed which uses the detected fault arc occurrence probability to represent the weight value of the algorithm, and combines the weight value and the detection results to make a comprehensive fault arc judgment.

[0125] In this example, step S16 may include:

[0126] (1) calculating the first detection result of the fault arc convolutional neural network detection by a softmax function to obtain a first weight value of the convolutional neural network detection result;

[0127] (2) calculating a second detection result of discrete wavelet transform detection by using an adaptive threshold value, and obtaining a second weight value of the discrete wavelet transform detection result;

[0128] (3) The first detection result and the first weight value are respectively combined with the second detection result and the second weight value, and an accurate detection result of the fault arc is obtained through weight arbitration.

[0129] The more severe the fault arc, the more high-frequency components the current signal contains, the greater the discreteness of the discrete wavelet transform result, the smaller the effect of the adjustment part in the adaptive threshold calculation formula, and the closer the threshold is to max / 2; when there is no fault arc or the fault arc is not severe, the current signal contains fewer high-frequency components, the discreteness of the discrete wavelet transform result is low, the adjustment part in the adaptive threshold calculation formula is effective, and the threshold will move up. Therefore, the ratio between half of the maximum value in the discrete wavelet transform result and the threshold is expressed as the first weight value. When the fault arc is severe, the first weight value w DWT The closer it is to 1, the less likely a fault arc will occur or the less severe the fault arc will occur. DWT The expression is shown in the following formula, where max is the maximum value in the detection result.

[0130]

[0131] In the multi-classification convolutional neural network algorithm, the softmax function is usually used to process the calculation results. Softmax can map the outputs of multiple neurons to the interval (0, 1) to represent the probability of each classification. The softmax function is divided into two steps. The first step is to use an exponential function to map the neural network results to the interval [0, +∞). The second step is to normalize all the results and divide the transformed results by the sum of all the transformed results to calculate the probability of each output as the second weight value.

[0132] The final result of the weighted arbitration algorithm is to add the first weight value of the discrete wavelet transform detection result and the second weight value of the convolutional neural network detection result. If the result is greater than 1, it is considered that the weighted arbitration algorithm determines that a fault arc has occurred. The weighted arbitration algorithm is used to determine the final result. A higher probability has a higher weight, which can effectively avoid false alarms.

[0133] After accurate detection of weight arbitration, an accurate judgment is made as to whether a fault arc occurs in the current signal of the cycle, and the first step can be returned to detect the fault current signal of the next cycle.

[0134] In the specific implementation of step S17, the number of fault cycles occurring within a unit time is counted according to the accurate detection result;

[0135] The detection results of the current signal are statistically analyzed. If the number of fault arcs detected within 1 second is greater than 14, it is considered that a fault arc has occurred; otherwise, it is considered that no fault arc has occurred.

[0136] In step S18, the quantity is compared with the national standard requirements to determine whether a fault arc occurs.

[0137] In the example of the present invention, the ADC sampling rate used for sampling is 1M / s, and the domestic power grid frequency is 50Hz, that is, the current waveform that can be collected within 1s is 100 half-sine wave cycles. According to the national standard, if the number of fault arcs detected within 1s is greater than 14, it is considered that a fault arc has occurred, that is, if more than 14 half-sine wave cycles are detected as fault arcs in every 100 half-sine wave cycles, it is considered that a fault arc has occurred in the load.

[0138] Compared with other neural network-implemented arc fault detection methods, the step-by-step fault arc detection method proposed in the embodiment of the present invention overcomes the false alarm situation in the convolutional neural network detection because it adopts the technical means of combining convolutional neural network detection with adaptive threshold discrete wavelet transform detection, thereby achieving the technical effect of further improving the accuracy rate. Because the pre-detection and precise detection step-by-step structure is adopted, the technical problem that the simplification and accuracy of the arc fault detection algorithm are difficult to coexist is overcome, thereby achieving the technical effect of simplifying the calculation process of the arc fault detection while ensuring high accuracy.

[0139] Corresponding to the above-mentioned embodiment of the step-by-step fault arc detection method, the present application also provides an embodiment of a step-by-step fault arc detection device.

[0140] Figure 4 FIG. 1 is a block diagram of a step-by-step fault arc detection device according to an exemplary embodiment. Figure 4 The device includes an acquisition module 11, an extraction module 12, a judgment module 13, a first judgment submodule 14, a second judgment submodule 15, a calculation module 16, a statistics module 17, and a comparison and judgment module 18.

[0141] An acquisition module 11, used to acquire a current signal to be detected;

[0142] An extraction module 12, used to extract time domain characteristic parameters from the current signal;

[0143] A judging module 13, used for judging the change value between the preceding and following consecutive cycles according to the time domain characteristic parameter;

[0144] A first judgment submodule 14 is used to consider that the detection result of the current cycle is the same as that of the previous cycle if the change value is less than a set threshold;

[0145] A second judgment submodule 15 is used for, if the change value is greater than a set threshold, detecting the current signal through a trained arc fault detection convolutional neural network to obtain a first detection result, detecting the current signal through discrete wavelet transform and comparing the detection result with an adaptive threshold to obtain a second detection result;

[0146] A calculation module 16, configured to calculate the first detection result by using a softmax function to obtain a first weight, calculate the second detection result by using an adaptive threshold to obtain a second weight, and obtain an accurate detection result by weight arbitration;

[0147] A statistical module 17, used for counting the number of fault cycles occurring within a unit time according to the precise detection result;

[0148] The comparison and judgment module 18 is used to compare the quantity with the national standard requirements to determine whether a fault arc occurs.

[0149] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0150] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0151] Correspondingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the step-by-step fault arc detection method as described above.

[0152] Correspondingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, characterized in that when the instructions are executed by a processor, the step-by-step fault arc detection method as described above is implemented.

[0153] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0154] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A step-by-step arc fault detection method, characterized in that: include: Acquiring a current signal to be detected; Extracting time domain characteristic parameters from the current signal; Determine the change value between the preceding and following consecutive cycles according to the time domain characteristic parameter; If the change value is less than the set threshold, it is considered that the detection result of the current cycle is the same as that of the previous cycle; If the change value is greater than the set threshold, the current signal is detected by a trained fault arc detection convolutional neural network to obtain a first detection result, and the current signal is detected by discrete wavelet transform and the detection result is compared with an adaptive threshold to obtain a second detection result; The first detection result is calculated by a softmax function to obtain a first weight, the second detection result is calculated by an adaptive threshold to obtain a second weight, and an accurate detection result is obtained through weight arbitration; According to the precise detection result, the number of fault cycles occurring within a unit time is counted; Compare the quantity with the national standard requirements to determine whether a fault arc occurs.

2. The method according to claim 1, characterized in that The current signal is an alternating current signal connected in series in the load circuit, and according to domestic standards, it should be 50Hz, 220V alternating current.

3. The step-by-step arc fault detection method according to claim 1, characterized in that: The time domain characteristic parameter is selected from one or more of a peak-to-peak value, a rectified mean value, an effective value and a variance.

4. The method according to claim 1, characterized in that: The construction method of the trained arc fault detection convolutional neural network is as follows: Acquire a current signal data set with an arc and a current signal data set without an arc; The convolutional neural network is trained by using the current signal data set with arc and the current signal data set without arc to obtain a trained fault arc detection convolutional neural network.

5. The method according to claim 1, characterized in that The arc fault detection convolutional neural network comprises: The first convolution layer is used to perform convolution calculation on the input current signal and extract the input current signal characteristics; The first maximum pooling layer is used to perform maximum pooling on the calculation results of the first convolutional layer and extract the signal features of the first convolutional layer; The second convolutional layer is used to perform convolution calculation on the result of the first maximum pooling layer to extract the features of the result of the first maximum pooling layer; The second maximum pooling layer is used to perform maximum pooling on the calculation results of the second convolutional layer and extract the signal features of the second convolutional layer; The third convolutional layer is used to perform convolution calculation on the result of the second maximum pooling layer and extract the features of the result of the second maximum pooling layer; The third maximum pooling layer is used to perform maximum pooling on the calculation results of the third convolutional layer and extract the signal features of the third convolutional layer; The tiling layer is used to combine and sort the calculation results of the third maximum pooling layer; The fully connected layer is used to further calculate the results of the tiled layer and output the calculation results of the corresponding labels as the first detection results.

6. The method according to claim 1, characterized in that Detecting the current signal by discrete wavelet transform and comparing the detection result with an adaptive threshold to obtain a second detection result, including: Using Db4 wavelet as mother wavelet, discrete wavelet transform is performed on the input fault arc current signal; The maximum value max of the statistical discrete wavelet transform result; Perform adaptive threshold calculation on discrete wavelet transform results; The second detection result is obtained by comparing the maximum value of the discrete wavelet transform detection result with the calculated adaptive threshold. If the maximum value of the discrete wavelet transform detection result is greater than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc; if the maximum value of the discrete wavelet transform detection result is less than the adaptive threshold, the second detection result is judged as the occurrence of a fault arc.

7. The method according to claim 1, characterized in that The first detection result is calculated by a softmax function to obtain a first weight, the second detection result is calculated by an adaptive threshold to obtain a second weight, and an accurate detection result is obtained by weight arbitration, including: Calculating the first detection result of the fault arc convolutional neural network detection by a softmax function to obtain a first weight value of the convolutional neural network detection result; Calculate the second detection result of discrete wavelet transform detection by using an adaptive threshold value, and obtain a second weight value of the discrete wavelet transform detection result; The first detection result and the first weight value are respectively combined with the second detection result and the second weight value, and an accurate detection result of the fault arc is obtained through weight arbitration.

8. A step-by-step arc fault detection device, characterized in that: include: An acquisition module, used for acquiring a current signal to be detected; An extraction module, used for extracting time domain characteristic parameters from the current signal; A judgment module, used for judging the change value between the preceding and following consecutive cycles according to the time domain characteristic parameter; A first judgment submodule, configured to consider that the detection result of the current cycle is the same as that of the previous cycle if the change value is less than a set threshold; A second judgment submodule is used for, if the change value is greater than a set threshold, detecting the current signal through a trained fault arc detection convolutional neural network to obtain a first detection result, detecting the current signal through discrete wavelet transform and comparing the detection result with an adaptive threshold to obtain a second detection result; A calculation module, configured to calculate the first detection result by using a softmax function to obtain a first weight, calculate the second detection result by using an adaptive threshold to obtain a second weight, and obtain an accurate detection result by weight arbitration; A statistical module, used for counting the number of fault cycles occurring within a unit time according to the precise detection result; The comparison and judgment module is used to compare the quantity with the national standard requirements to determine whether a fault arc occurs.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images