DC series arc fault detection method and device, photovoltaic inverter and medium

Abstract

The invention relates to the technical field of direct current fault detection, and discloses a direct current series arc fault detection method and device, a photovoltaic inverter and a medium. And processing the input voltage and the input current of the photovoltaic inverter to obtain a characteristic data training set under the condition that the arc generator simulates the photovoltaic inverter to generate a direct current series arc fault and the condition of normal operation, wherein each arc fault detection model is obtained by pre-training based on the corresponding characteristic data set. The method comprises the following steps: firstly, acquiring an input side voltage and an input side current of the photovoltaic inverter, and processing the input side voltage and the input side current respectively to obtain multiple groups of feature data including voltage feature parameters and current feature parameters; then, the multiple sets of feature data are input into a pre-trained corresponding arc fault detection model for processing, and multiple fault detection results are obtained. And finally, based on a plurality of fault detection results, a voting mechanism is adopted to determine whether the photovoltaic inverter has a DC series arc fault.

Description

Technical Field

[0001] This invention relates to the field of DC fault detection technology, and in particular to a DC series arc fault detection method, device, photovoltaic inverter and dielectric. Background Technology

[0002] As the photovoltaic industry continues to expand, related safety hazards are gradually becoming more prominent. Among them, DC series arc faults seriously affect the normal operation of photovoltaic systems, so detecting DC series arc faults is crucial.

[0003] In related technologies, DC series arc fault detection is achieved by utilizing sensors to receive the physical characteristics of arc light, heat, and electromagnetic radiation during arc discharge. However, these methods require high sensor accuracy, which limits the detection range. Therefore, a new method for DC series arc fault detection is needed. Summary of the Invention

[0004] The embodiments described in this specification aim to at least partially solve one of the technical problems in the related art. To this end, the embodiments described in this specification propose a method, apparatus, photovoltaic inverter, and dielectric for detecting DC series arc faults.

[0005] This specification provides a method for detecting DC series arc faults, applied to photovoltaic inverters. The method includes:

[0006] The input voltage and input current of the photovoltaic inverter are obtained, and the input voltage and input current are processed respectively to obtain multiple sets of feature data, wherein each set of feature data includes voltage feature parameters and current feature parameters;

[0007] The multiple sets of feature data are respectively input into the corresponding pre-trained arc fault detection models for processing to obtain multiple fault detection results. Each arc fault detection model is pre-trained based on the corresponding feature dataset. Each feature data training set is obtained by processing the input voltage and input current of the photovoltaic inverter under the conditions of the arc generator simulating the DC series arc fault of the photovoltaic inverter and the normal operation of the photovoltaic inverter.

[0008] Based on the multiple fault detection results, a voting mechanism is used to determine whether the photovoltaic inverter has experienced a DC series arc fault.

[0009] In one embodiment, the input-side voltage is processed, including:

[0010] The input voltage is subjected to voltage following and bandpass filtering, and the processed voltage value is calculated based on a time window to obtain the voltage value within the time window;

[0011] The voltage values ​​within the time window are subjected to time-domain and frequency-domain feature extraction to obtain voltage time-domain feature parameters and voltage frequency-domain feature parameters.

[0012] In one embodiment, the voltage time-domain characteristic parameters include voltage kurtosis, voltage waveform factor, and voltage margin factor, and the voltage frequency-domain characteristic parameters include voltage spectrum integrals over a preset frequency band.

[0013] In one embodiment, after obtaining the voltage time-domain characteristic parameters and the voltage frequency-domain characteristic parameters, the method further includes:

[0014] The voltage time-domain characteristic parameters and voltage frequency-domain characteristic parameters are normalized.

[0015] In one embodiment, the input-side current is processed, including:

[0016] The input current is filtered and amplified, and the processed current value is calculated based on a time window to obtain the current value within the time window.

[0017] The current value within the time window is subjected to time-domain features, frequency-domain features, and time-frequency-domain features to obtain current time-domain feature parameters, current frequency-domain feature parameters, and current time-frequency-domain feature parameters.

[0018] Principal component analysis is used to reduce the dimensionality of the current time-domain characteristic parameters, the current frequency-domain characteristic parameters, and the current time-frequency-domain characteristic parameters.

[0019] In one embodiment, the current time-domain characteristic parameters include current kurtosis, current waveform factor, and current margin factor; the current frequency-domain characteristic parameters include the current spectrum integral of a preset frequency band; and the current time-frequency-domain characteristic parameters include the modulus maxima and variance of the wavelet coefficients of the sub-frequency bands after three-level decomposition of the wavelet packet.

[0020] In one embodiment, after performing principal component analysis dimensionality reduction on the current time-domain feature parameters, the current frequency-domain feature parameters, and the current time-frequency-domain feature parameters, the method further includes:

[0021] The current characteristic parameters obtained by dimensionality reduction are normalized.

[0022] In one implementation, based on the multiple fault detection results, a voting mechanism is used to determine whether the photovoltaic inverter has experienced a DC series arc fault, including:

[0023] If, based on the multiple fault detection results, at least two arc fault detection models detect an arc fault in the photovoltaic inverter, it is determined that the photovoltaic inverter has experienced a DC series arc fault.

[0024] This specification provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the above embodiments.

[0025] This specification provides a computer program product that includes instructions that, when executed by a processor of a computer device, enable the computer device to perform the steps of the method described in any of the above embodiments.

[0026] This specification provides a photovoltaic inverter, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the DC series arc fault detection method according to any of the above embodiments.

[0027] This specification provides a DC series arc fault detection device for use in a photovoltaic inverter. The device includes:

[0028] The processing module is used to acquire the input side voltage and input side current of the photovoltaic inverter, and process the input side voltage and input side current respectively to obtain multiple sets of feature data, wherein each set of feature data includes voltage feature parameters and current feature parameters;

[0029] The fault detection module is used to input the multiple sets of feature data into the corresponding pre-trained arc fault detection models for processing, obtain multiple fault detection results, and use a voting mechanism to determine whether the photovoltaic inverter has experienced a DC series arc fault based on the multiple fault detection results. Each arc fault detection model is pre-trained based on a corresponding feature dataset. Each feature data training set is obtained by processing the input voltage and input current of the photovoltaic inverter under the conditions of an arc generator simulating a DC series arc fault in the photovoltaic inverter and under the condition of normal operation of the photovoltaic inverter.

[0030] The above-described implementation method is applied to photovoltaic inverters. Before detecting DC-DC series arc faults, the input voltage and current of the photovoltaic inverter are processed to obtain a feature data training set under conditions simulating a DC-DC series arc fault in the photovoltaic inverter using an arc generator and under conditions of normal operation of the photovoltaic inverter. Each arc fault detection model is pre-trained based on the corresponding feature dataset. During the DC-DC series arc fault detection process, firstly, the input side voltage and input side current of the photovoltaic inverter are acquired and processed separately to obtain multiple sets of feature data, including voltage feature parameters and current feature parameters. Then, the multiple sets of feature data are input into the pre-trained corresponding arc fault detection models for processing to obtain multiple fault detection results. Finally, based on the multiple fault detection results, a voting mechanism is used to determine whether a DC-DC series arc fault has occurred in the photovoltaic inverter. In the above embodiments, multiple feature data training sets obtained by processing the input voltage and input current of the photovoltaic inverter under conditions of DC series arc fault in the photovoltaic inverter simulated by an arc generator and under conditions of normal operation of the photovoltaic inverter are used to train the corresponding arc fault detection model. This makes the DC series arc fault detection method applicable to multiple scenarios, improving the universality and robustness of the method for detecting DC series arc faults in practical applications. Using multiple sets of arc fault detection models avoids the drawback of a single model being prone to misjudgment, improving the accuracy of arc fault detection. A voting mechanism is adopted to determine whether a DC series arc fault has occurred in the photovoltaic inverter based on multiple fault detection results, improving the reliability and accuracy of DC series arc fault detection. Attached Figure Description

[0031] Figure 1 A schematic diagram of the online monitoring circuit for the DC series arc fault detection method provided in the embodiments of this specification;

[0032] Figure 2a A flowchart illustrating the DC series arc fault detection method provided in the embodiments of this specification;

[0033] Figure 2b A circuit diagram of the training arc fault detection model provided for the embodiments of this specification;

[0034] Figure 3 A schematic diagram illustrating the results of the real and predicted classes provided for the implementation of this specification;

[0035] Figure 4 A schematic diagram illustrating the results of the real and predicted classes provided for the implementation of this specification;

[0036] Figure 5 A schematic flowchart illustrating the processing of the input-side voltage provided for embodiments of this specification;

[0037] Figure 6 A flowchart illustrating the processing of input-side current provided for embodiments of this specification;

[0038] Figure 7 A flowchart illustrating the DC series arc fault detection method provided in the embodiments of this specification;

[0039] Figure 8 A schematic diagram of a DC series arc fault detection device provided for embodiments of this specification. Detailed Implementation

[0040] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0041] Constructing a new power system based primarily on new energy sources such as solar and wind power is of great significance. However, the DC series arc in photovoltaic power generation systems differs from AC and parallel arcs. A DC series arc lacks a zero-crossing point and is difficult to extinguish on its own. Furthermore, the current amplitude change during a DC series arc is minimal, making it difficult to detect and thus hard to extinguish, potentially causing fires and hindering the development of photovoltaic systems.

[0042] In related technologies, the detection of DC series arc faults is achieved by using sensors to receive the physical characteristics of arc light, heat, and electromagnetic radiation during arc discharge. However, this method requires high sensor accuracy and is greatly affected by complex environments, thus limiting its detection range.

[0043] In related technologies, DC series arc fault detection is performed using the electrical characteristics of electric arcs, such as the time-frequency domain features of current and voltage. However, this method of detection based on the electrical characteristics of electric arcs is susceptible to false alarms due to interference from internal and external noise sources within the photovoltaic system.

[0044] In related technologies, machine learning algorithms are used to extract fault feature information to achieve DC series arc fault detection. However, the detection method using machine learning algorithms requires a lot of time to extract features, and its performance is affected by the number and quality of samples, while also having high hardware requirements.

[0045] Based on the above analysis, this specification provides a method for detecting DC series arc faults. This method is applied to photovoltaic inverters. Before detecting DC series arc faults, the input voltage and current of the photovoltaic inverter are processed under conditions simulating a DC series arc fault in the inverter using an arc generator, as well as under conditions of normal operation of the photovoltaic inverter, to obtain a feature data training set. Each arc fault detection model is pre-trained based on the corresponding feature dataset. During the DC series arc fault detection process, firstly, the input side voltage and input side current of the photovoltaic inverter are acquired and processed separately to obtain multiple sets of feature data, including voltage and current feature parameters. Then, the multiple sets of feature data are input into the pre-trained corresponding arc fault detection models for processing to obtain multiple fault detection results. Finally, based on the multiple fault detection results, a voting mechanism is used to determine whether a DC series arc fault has occurred in the photovoltaic inverter. In the above embodiments, multiple feature data training sets obtained by processing the input voltage and input current of the photovoltaic inverter under conditions of DC series arc fault in the photovoltaic inverter simulated by an arc generator and under conditions of normal operation of the photovoltaic inverter are used to train the corresponding arc fault detection model. This makes the DC series arc fault detection method applicable to multiple scenarios, improving the universality and robustness of the method for detecting DC series arc faults in practical applications. Using multiple sets of arc fault detection models avoids the drawback of a single model being prone to misjudgment, improving the accuracy of arc fault detection. A voting mechanism is adopted to determine whether a DC series arc fault has occurred in the photovoltaic inverter based on multiple fault detection results, improving the reliability and accuracy of DC series arc fault detection.

[0046] Please refer to the online monitoring circuit diagram corresponding to the DC series arc fault detection method provided in this specification. Figure 1 The circuit diagram includes a photovoltaic inverter 102, a photovoltaic cell 104, a line impedance 106, a current transformer 108, a signal sampling and processing module 110, and a fault detection module 112. The line impedance 106 simulates the line impedance from the photovoltaic module to the inverter in a real-world scenario. The current transformer 108 collects the line current from the photovoltaic cell 104 to the photovoltaic inverter 102 in real time under various conditions.

[0047] Specifically, the voltage acquisition circuit within the signal sampling and processing module 110 acquires the line voltage signal data between the photovoltaic cell 104 and the photovoltaic inverter 102 in real time to obtain the input-side voltage of the photovoltaic inverter 102. Similarly, the current acquisition circuit within the signal sampling and processing module 110 acquires the line current signal data between the photovoltaic cell 104 and the photovoltaic inverter 102 in real time to obtain the input-side current of the photovoltaic inverter 102. The input-side voltage is then processed by a voltage follower circuit and a bandpass filter circuit within the signal sampling and processing module 110 to obtain the processed voltage value. The input-side current is then filtered and amplified by a bandpass filter circuit and an amplification circuit within the signal sampling and processing module 110 to obtain the processed current value. When the sampling time for the real-time acquisition of the input-side current and input-side voltage of the photovoltaic inverter 102 reaches a time window, the processed voltage value is calculated based on the time window to obtain the voltage value within the time window, and the processed current value is calculated based on the time window to obtain the current value within the time window. Then, the microprocessor in the fault detection module 112 extracts time-domain and frequency-domain features from the voltage values ​​within the time window to obtain voltage time-domain and voltage frequency-domain feature parameters. The microprocessor in the fault detection module 112 also extracts time-domain, frequency-domain, and time-frequency-domain features from the current values ​​within the time window to obtain current time-domain, current frequency-domain, and current time-frequency-domain feature parameters. Next, the microprocessor in the fault detection module 112 performs principal component analysis (PCA) dimensionality reduction on the current time-domain, current frequency-domain, and current time-frequency-domain feature parameters to obtain dimensionality-reduced current feature parameters. The microprocessor in the fault detection module 112 normalizes the voltage time-domain and voltage frequency-domain feature parameters, and normalizes the dimensionality-reduced current feature parameters to obtain five sets of feature data. Finally, the microprocessor in the fault detection module 112 inputs these five sets of feature data into pre-trained arc fault detection models for processing, obtaining five fault detection results. Of the five fault detection results, four indicate that the corresponding arc fault detection model has detected an arc fault in the photovoltaic inverter, confirming a DC series arc fault in the photovoltaic inverter. If an arc fault is detected, the alarm module (e.g., a buzzer) within the fault detection module 112 will sound an alarm, and the display module (e.g., an indicator light) will change from green to red.

[0048] This specification provides a method for detecting DC series arc faults, applied to photovoltaic inverters. Please refer to [link / reference]. Figure 2a The DC series arc fault detection method may include the following steps:

[0049] S210. Obtain the input voltage and input current of the photovoltaic inverter, and process the input voltage and input current respectively to obtain multiple sets of characteristic data.

[0050] Each set of characteristic data includes voltage characteristic parameters and current characteristic parameters. The input-side voltage can be the line voltage flowing into the photovoltaic inverter, for example, the line voltage between the photovoltaic module and the photovoltaic inverter. The input-side current can be the line current flowing into the photovoltaic inverter, for example, the line current between the photovoltaic module and the photovoltaic inverter.

[0051] Specifically, voltage sensors and circuits are used to collect the line voltage flowing into the photovoltaic inverter, obtaining the input voltage. This input voltage contains noise and interference; signal conditioning and feature extraction are performed to process the input voltage and obtain voltage characteristic parameters. Similarly, current sensors, circuits, and current transformers are used to collect the line current flowing into the photovoltaic inverter, obtaining the input current. This input current also contains noise; signal conditioning and feature extraction are performed to process the input current and obtain current characteristic parameters. Combining different voltage and current characteristic parameters can create multiple sets of feature data.

[0052] S220. Input multiple sets of feature data into the corresponding pre-trained arc fault detection models for processing to obtain multiple fault detection results.

[0053] Each arc fault detection model is pre-trained based on a corresponding feature dataset. Each feature dataset is obtained by processing the input voltage and current of the photovoltaic inverter under both DC series arc fault conditions simulated by the arc generator and normal operation conditions. The input voltage can be the line voltage flowing into the photovoltaic inverter under different operating conditions simulated by the arc generator; for example, the input voltage could be the line voltage between the photovoltaic module and the photovoltaic inverter. Similarly, the input current can be the line current flowing into the photovoltaic inverter under different operating conditions simulated by the arc generator; for example, the input current could be the line current between the photovoltaic module and the photovoltaic inverter.

[0054] Specifically, multiple sets of feature data are used as input data for a pre-trained arc fault detection model, with each set of feature data corresponding to a pre-trained arc fault detection model. The arc fault detection model corresponding to each set of feature data processes the set of feature data, extracts the relationships between features, generates prediction results, and outputs fault detection results based on the prediction results. Since there are multiple sets of feature data, multiple fault detection results can be obtained. The arc fault detection model can be built based on a BP neural network (Backpropagation Neural Network).

[0055] It should be noted that you should refer to [link / reference]. Figure 2b , Figure 2b The circuit diagram for training the arc fault detection model is shown. The circuit diagram includes a photovoltaic inverter 202, photovoltaic cells 204, line impedance 206, current transformer 208, arc generator 210, signal sampling and processing module 212, and fault detection module 214. Line impedance 206 simulates the line impedance from the photovoltaic module to the inverter in a real-world scenario. Current transformer 208 collects the line current from photovoltaic cells 204 to photovoltaic inverter 202 in real time under various conditions.

[0056] Specifically, the photovoltaic inverter is started and allowed to operate normally without the intervention of an arc generator. Current transformer 208 collects the line current signal data between photovoltaic cell 204 and photovoltaic inverter 202 to obtain the input current of photovoltaic inverter 202 under conditions where there is no DC series arc fault. Similarly, voltage transformer in signal sampling and processing module 212 collects the line voltage signal data between photovoltaic cell 204 and photovoltaic inverter 202 to obtain the input voltage of photovoltaic inverter 202 under conditions where there is no DC series arc fault. After the system is running stably, an arc is introduced into the DC circuit using an arc generator to simulate the generation of an arc fault. By adjusting the parameters of the arc generator, arcs of different intensities are generated, thereby simulating arc faults of varying degrees. The arc generator generates an arc according to the adjusted parameters. The current transformer 208 collects the line current signal data between the photovoltaic cell 204 and the photovoltaic inverter 202 to obtain the input current when the photovoltaic inverter 202 experiences a DC series arc fault. The voltage transformer in the signal sampling and processing module 212 collects the line voltage signal data between the photovoltaic cell 204 and the photovoltaic inverter 202 to obtain the input voltage when the photovoltaic inverter 202 experiences a DC series arc fault.

[0057] The input current of the photovoltaic inverter 202 under both DC series arc fault and non-DC series arc fault conditions is filtered and amplified by the bandpass filter and amplification circuit in the signal sampling and processing module 212 to obtain the processed current value. Similarly, the input voltage of the photovoltaic inverter 202 under both DC series arc fault and non-DC series arc fault conditions is processed by the voltage follower circuit and bandpass filter in the signal sampling and processing module 212 to obtain the processed voltage value. When the sampling time for the real-time acquisition of the input current and input voltage of the photovoltaic inverter 102 reaches the time window, the processed voltage value is calculated based on the time window to obtain the voltage value within the time window, and the processed current value is calculated based on the time window to obtain the current value within the time window. Then, the microprocessor in the fault detection module 214 extracts the time-domain and frequency-domain features of the voltage value within the time window to obtain the voltage time-domain feature parameters and voltage frequency-domain feature parameters. The microprocessor in the fault detection module 214 extracts time-domain, frequency-domain, and time-frequency-domain features from the current values ​​within the time window, obtaining current time-domain feature parameters, current frequency-domain feature parameters, and current time-frequency-domain feature parameters. Then, the microprocessor in the fault detection module 214 performs principal component analysis (PCA) to reduce the dimensionality of the current time-domain, current frequency-domain, and current time-frequency-domain feature parameters, obtaining the dimensionality-reduced current feature parameters. The microprocessor in the fault detection module 214 normalizes the voltage time-domain and voltage frequency-domain feature parameters, and normalizes the dimensionality-reduced current feature parameters, forming m (m≥2) sets of feature data training sets. These m sets of feature data training sets are labeled. Each labeled set of feature data training sets is then used as input to the corresponding arc fault detection model to be trained, yielding the corresponding predicted fault result. Based on the label of each set of feature data training sets and the corresponding predicted fault result, the loss value of the arc fault detection model to be trained can be determined. The parameters of the arc fault detection model to be trained are then updated based on the loss value. This process continues, training the updated arc fault detection model until the training stops, at which point a fully trained arc fault detection model is obtained. Since there are m sets of feature data training sets, different sets can be used for different arc fault detection models, and different combinations of these sets can also be used to train different models. Therefore, through this implementation method, by training the arc fault detection model using different feature data training sets, n fully trained arc fault detection models can be obtained and saved for use in practical applications. The training stopping condition can be either the model loss value converging or the number of training epochs reaching a preset number.

[0058] It should be noted that a suitable arc generator can be selected to simulate DC series arc faults based on the simulation requirements and characteristics, and this application does not impose any restrictions on this.

[0059] S230. Based on multiple fault detection results, a voting mechanism is used to determine whether a DC series arc fault has occurred in the photovoltaic inverter.

[0060] Among them, the voting mechanism can be a method for summarizing and comprehensively evaluating multiple fault detection results.

[0061] Specifically, a voting rule is defined to determine how many arc fault detection models are needed to detect a DC-DC series arc fault before the photovoltaic inverter is considered to have experienced a DC-DC series arc fault. Specifically, the judgment is made based on the fault detection results of multiple arc fault detection models to determine whether each model detected an arc fault in the photovoltaic inverter. Then, using the voting mechanism, the results of each arc fault detection model's detection of an arc fault in the photovoltaic inverter are used to determine whether a DC-DC series arc fault has occurred in the photovoltaic inverter.

[0062] In the aforementioned DC series arc fault detection method, applied to photovoltaic inverters, before detecting DC series arc faults, the input voltage and current of the photovoltaic inverter are processed under conditions simulating a DC series arc fault in the photovoltaic inverter and under conditions of normal operation of the photovoltaic inverter to obtain a feature data training set. Each arc fault detection model is pre-trained based on the corresponding feature dataset. During the DC series arc fault detection process, firstly, the input side voltage and input side current of the photovoltaic inverter are acquired and processed separately to obtain multiple sets of feature data including voltage and current feature parameters. Then, these multiple sets of feature data are input into the pre-trained corresponding arc fault detection models for processing, obtaining multiple fault detection results. Finally, based on the multiple fault detection results, a voting mechanism is used to determine whether a DC series arc fault has occurred in the photovoltaic inverter. In the above embodiments, multiple feature data training sets obtained by processing the input voltage and input current of the photovoltaic inverter under conditions of DC series arc fault in the photovoltaic inverter simulated by an arc generator and under conditions of normal operation of the photovoltaic inverter are used to train the corresponding arc fault detection model. This makes the DC series arc fault detection method applicable to multiple scenarios, improving the universality and robustness of the method for detecting DC series arc faults in practical applications. Using multiple sets of arc fault detection models avoids the drawback of a single model being prone to misjudgment, improving the accuracy of arc fault detection. A voting mechanism is adopted to determine whether a DC series arc fault has occurred in the photovoltaic inverter based on multiple fault detection results, improving the reliability and accuracy of DC series arc fault detection.

[0063] In some implementations, there are multiple pre-trained arc fault detection models. These pre-trained arc fault detection models can be trained using a training set of feature data obtained by normalizing voltage time-domain feature parameters and voltage frequency-domain feature parameters.

[0064] It should be noted that after successfully training the arc fault detection model, the accuracy of the model's output results is analyzed. After training the arc fault detection model, an arc generator is used to simulate both DC series arc faults and non-DC series arc faults in the photovoltaic inverter. 500 sets of input voltages are obtained for the case without DC series arc faults and 500 sets for the case with DC series arc faults. Voltage tracking and bandpass filtering, time-domain and frequency-domain feature extraction, and normalization are then performed on these input voltages. The normalized results are labeled to obtain the test dataset. The test dataset is input into the trained arc fault detection model, which processes the dataset to obtain predicted data. The labels are set to 1 and 0, where 1 indicates the presence of a DC series arc fault in the photovoltaic inverter, and 0 indicates the absence of a DC series arc fault.

[0065] Please see Figure 3 The true class is the label corresponding to the test dataset, and the predicted class is the predicted data obtained through the arc fault detection model. The probability that the label corresponding to the test dataset is 1 and the predicted data obtained through the arc fault detection model is also 1 is 99.5%; the probability that the label corresponding to the test dataset is 0 and the predicted data obtained through the arc fault detection model is also 0 is 99.7%; the probability that the label corresponding to the test dataset is 1 and the predicted data obtained through the arc fault detection model is also 0 is 0.5%; and the probability that the label corresponding to the test dataset is 0 and the predicted data obtained through the arc fault detection model is also 1 is 0.3%.

[0066] Furthermore, the arc fault detection model A trained using a training set of feature data obtained by normalizing voltage time-domain and voltage frequency-domain feature parameters exhibits better performance than arc detection models trained using random forest, support vector machine, and Naive Bayes methods. The formulas for calculating accuracy, precision, and recall are shown below:

[0067]

[0068] Where TP represents the number of positive examples correctly predicted by the model, TN represents the number of negative examples correctly predicted by the model, FP represents the number of positive examples incorrectly predicted by the model, and FN represents the number of negative examples incorrectly predicted by the model.

[0069] The results are shown in Table 1:

[0070]

[0071] Table 1

[0072] In some implementations, there are multiple pre-trained arc fault detection models. Among them, the pre-trained arc fault detection models can be trained based on data obtained by normalizing the current feature parameters obtained by dimensionality reduction processing.

[0073] It should be noted that after successfully training the arc fault detection model, the accuracy of the model's output results is analyzed. After training the arc fault detection model, an arc generator is used to simulate both DC series arc faults and non-DC series arc faults in the photovoltaic inverter. 500 sets of input currents are obtained for the case without DC series arc faults and 500 sets for the case with DC series arc faults. These are then filtered, amplified, and subjected to time-domain, frequency-domain, and time-frequency-domain feature extraction and normalization. The normalized results are labeled to obtain the test dataset. The test dataset is input into the trained arc fault detection model, which processes the dataset to obtain predicted data. The labels are set to 1 and 0, where 1 indicates the presence of a DC series arc fault in the photovoltaic inverter, and 0 indicates the absence of a DC series arc fault.

[0074] Please see Figure 4 The true class is the label corresponding to the test dataset, and the predicted class is the predicted data obtained through the arc fault detection model. The probability that the label corresponding to the test dataset is 1 and the predicted data obtained through the arc fault detection model is also 1 is 98.9%; the probability that the label corresponding to the test dataset is 0 and the predicted data obtained through the arc fault detection model is also 0 is 99.4%; the probability that the label corresponding to the test dataset is 1 and the predicted data obtained through the arc fault detection model is also 0 is 1.1%; and the probability that the label corresponding to the test dataset is 0 and the predicted data obtained through the arc fault detection model is also 1 is 0.6%.

[0075] Furthermore, the arc fault detection model B, trained using data obtained by normalizing the current feature parameters obtained through dimensionality reduction, exhibits better performance than arc detection models trained using random forest, support vector machine, and BP neural network methods. The formulas for calculating accuracy, precision, and recall are shown below:

[0076]

[0077]

[0078] Where TP represents the number of positive examples correctly predicted by the model, TN represents the number of negative examples correctly predicted by the model, FP represents the number of positive examples incorrectly predicted by the model, and FN represents the number of negative examples incorrectly predicted by the model.

[0079] The results are shown in Table 2:

[0080]

[0081] Table 2

[0082] In some implementations, please refer to Figure 5 Processing the input voltage may include the following steps:

[0083] S510 performs voltage following and bandpass filtering on the input voltage, and calculates the processed voltage value based on the time window to obtain the voltage value within the time window.

[0084] S520. Extract time-domain and frequency-domain features from the voltage values ​​within the time window to obtain voltage time-domain feature parameters and voltage frequency-domain feature parameters.

[0085] Voltage following processing can adjust the input voltage according to a given reference signal to make it follow the changes of the reference signal. Bandpass filtering processing can filter the input voltage, allowing input voltages within a specified frequency range to pass through.

[0086] Specifically, due to various noises and interferences in actual circuits, the acquired input voltage contains noise. A voltage follower is used to perform voltage following processing on the input voltage, resulting in an output voltage that is identical to the input voltage but has a stronger current driving capability. Then, a bandpass filter is applied to the output voltage, allowing output voltages within a certain frequency range to pass through, yielding the processed voltage value. The acquisition time of the input voltage is calculated. When the acquisition time reaches the time window, the voltage value within the time window is obtained based on the processed voltage value within that window. Time-domain features (such as root mean square value, waveform factor, and margin factor) are extracted from the voltage value within the time window to obtain voltage time-domain feature parameters describing the basic statistical characteristics of the voltage waveform. A Fourier transform is performed on the voltage value within the time window to convert it from the time domain to the frequency domain. The transformed spectrum is analyzed, and the amplitude of each frequency component is calculated. Frequency-domain features are extracted from the spectrum to obtain voltage frequency-domain feature parameters describing the distribution characteristics of the voltage signal in the frequency domain.

[0087] In some implementations, outliers are defined based on business requirements and signal characteristics, such as data points exceeding a certain range. Statistical methods, threshold-based methods, etc., can be used to detect outliers in the voltage values ​​after voltage tracking and bandpass filtering of the input voltage. Detected outliers are removed from the processed voltage values ​​to ensure the accuracy and reliability of subsequent analysis, resulting in the removed voltage values. The removed voltage values ​​are then calculated based on a time window to obtain the voltage values ​​within that time window.

[0088] In the aforementioned DC series arc fault detection method, the input voltage is subjected to voltage tracking and bandpass filtering. The processed voltage value is then calculated based on a time window to obtain the voltage value within that time window. This reduces interference from the external environment and noise on the input voltage, improving the accuracy of DC fault detection and avoiding false positives. Furthermore, time-domain and frequency-domain features are extracted from the voltage value within the time window to obtain voltage time-domain and frequency-domain feature parameters, avoiding the influence of a single feature parameter on the accuracy of DC fault detection.

[0089] In some implementations, the voltage time-domain characteristic parameters include voltage kurtosis, voltage waveform factor, and voltage margin factor, while the voltage frequency-domain characteristic parameters include the voltage spectrum integral of a preset frequency band.

[0090] Among these, voltage kurtosis can be a statistical measure describing the shape of a voltage signal waveform, measuring the degree to which a data value deviates from the average value. Voltage waveform factor can be a statistical parameter describing the shape of a voltage waveform, measuring the difference between the actual voltage waveform and an ideal waveform (such as a sine wave). Voltage margin factor can be a parameter measuring the fluctuation range or volatility of a voltage signal. Voltage spectrum integral can be a characteristic parameter describing the energy distribution of a voltage signal in the frequency domain.

[0091] Specifically, to avoid the problem that dimensional parameters are sensitive to voltage characteristics and easily affected by interference, dimensionless parameters such as voltage kurtosis, voltage waveform factor, and voltage margin factor are selected as voltage time-domain characteristic parameters. In the frequency domain, a comprehensive measure is needed to describe the frequency distribution of the voltage signal, while not depending on specific frequency values ​​and reducing the sensitivity of parameters to voltage changes. Voltage spectrum integral can be used as the voltage frequency-domain characteristic parameter.

[0092] For example, the formulas for voltage kurtosis, voltage waveform factor, and voltage margin factor are obtained by extracting time-domain features from the voltage values ​​within the time window, as shown below:

[0093]

[0094] Where ku(f) is the voltage kurtosis, F(f) is the voltage waveform factor, and M(f) is the voltage margin factor. N is the number of sampling data points within the time window, and f i f represents the amplitude of the voltage sampling data points within the time window. max The maximum amplitude of the voltage data points within the time window. This represents the average amplitude of the sampled data points of the voltage value within the time window.

[0095] The frequency spectrum integral within the range of 30kHz to 125kHz after Fourier transform can be used as a voltage frequency domain characteristic parameter. The frequency spectrum integral is determined according to the following formula:

[0096]

[0097] Where C(m) is the spectral integral, i.e., the voltage frequency domain characteristic parameter, N is the number of sampling data points in the time window, f1 represents the frequency band, f2 represents the frequency band, and m is the spectral amplitude of the voltage value in the time window after Fourier transform. It can eliminate the influence of the number of sampling data points on the voltage frequency domain characteristics.

[0098] In the above-mentioned DC series arc fault detection method, the voltage time-domain characteristic parameters include voltage kurtosis, voltage waveform factor and voltage margin factor, and the voltage frequency-domain characteristic parameters include voltage spectrum integral of preset frequency band, so as to understand the characteristics of current more comprehensively and accurately and improve the accuracy of DC series arc fault detection.

[0099] In some implementations, after obtaining the voltage time-domain characteristic parameters and the voltage frequency-domain characteristic parameters, the method may further include: normalizing the voltage time-domain characteristic parameters and the voltage frequency-domain characteristic parameters.

[0100] Specifically, in the voltage time-domain and voltage frequency-domain feature parameters extracted from time-domain and frequency-domain features, each voltage time-domain feature parameter and voltage frequency-domain feature parameter may have different dimensions and numerical ranges. To eliminate these differences, it is necessary to normalize (e.g., standardize, scale) the voltage time-domain and voltage frequency-domain feature parameters so that they have the same magnitude, thereby ensuring that the influence of each feature on the final result is equal during the training of the arc fault detection model.

[0101] In the above-mentioned DC series arc fault detection method, the voltage time-domain characteristic parameters and voltage frequency-domain characteristic parameters are normalized to improve the performance of the arc fault detection model.

[0102] In some implementations, please refer to Figure 6 Processing the input-side current may include the following steps:

[0103] S610: Filter and amplify the input current, and calculate the processed current value based on the time window to obtain the current value within the time window.

[0104] S620. Extract time-domain, frequency-domain, and time-frequency-domain features from the current value within the time window to obtain current time-domain feature parameters, current frequency-domain feature parameters, and current time-frequency-domain feature parameters.

[0105] S630. Principal component analysis is used to reduce the dimensionality of the current time-domain characteristic parameters, current frequency-domain characteristic parameters, and current time-frequency-domain characteristic parameters.

[0106] Specifically, due to various noises and interferences in actual circuits, the acquired input current contains noise. Therefore, filters such as low-pass, band-pass, or high-pass filters are used to filter the input current, resulting in a filtered current. The filtered current may be weak, so an operational amplifier or other type of amplifier is used to amplify it, yielding a processed current value. The acquisition time of the input current is calculated. When the acquisition time reaches the time window, the current value within the time window is obtained based on the processed current value within that window. Time-domain features (such as root mean square value, waveform factor, and margin factor) are extracted from the current value within the time window to obtain current time-domain feature parameters describing the basic statistical characteristics of the current waveform. A Fourier transform is performed on the current value within the time window to convert it from the time domain to the frequency domain. The transformed spectrum is analyzed, and the amplitude of each frequency component is calculated. Frequency-domain features are extracted from the spectrum to obtain current frequency-domain feature parameters describing the distribution characteristics of the current signal in the frequency domain. The current values ​​within a time window are processed using time-frequency analysis techniques such as Fourier transform and wavelet transform. Time-frequency domain features are then extracted from the results to obtain the current time-frequency domain feature parameters. In some implementations, a dimensionality reduction model can be used to perform principal component analysis (PCA) on the current time-domain, current frequency-domain, and current time-frequency domain feature parameters. This projects the current time-domain, current frequency-domain, and current time-frequency domain feature parameters into a new feature space, thereby reducing the data dimensionality while retaining most of the data information, resulting in the dimensionality-reduced current feature parameters. In other implementations, the current time-domain, current frequency-domain, and current time-frequency domain feature parameters are standardized to have the same mean and variance to avoid significant impacts from differences between feature parameters on the PCA results. Then, a dimensionality reduction model is used to perform PCA on the standardized current time-domain, current frequency-domain, and current time-frequency domain feature parameters to obtain the dimensionality-reduced current feature parameters.

[0107] In other implementations, outliers are defined based on business requirements and signal characteristics, such as data points exceeding a certain range. Statistical methods or threshold-based methods can be used to filter and amplify the input current to obtain processed current values. Outlier detection is then performed on these processed current values, removing detected outliers to ensure the accuracy and reliability of subsequent analysis, resulting in the removed current values. The removed current values ​​are then calculated based on a time window to obtain the current values ​​within that time window.

[0108] In the aforementioned DC series arc fault detection method, the input current is filtered and amplified, and the processed current value is calculated based on a time window to obtain the current value within the time window. This reduces the interference of external environment and noise on the input current, improves the accuracy of DC fault detection, and avoids false positives. Time-domain, frequency-domain, and time-frequency-domain features are extracted from the current value within the time window to obtain current time-domain feature parameters, current frequency-domain feature parameters, and current time-frequency-domain feature parameters. Principal component analysis is then performed on these current time-domain, current frequency-domain, and current time-frequency-domain feature parameters to reduce dimensionality. Simultaneously considering the time-domain, frequency-domain, and time-frequency-domain features of the current avoids the influence of a single feature parameter on the accuracy of DC fault detection.

[0109] In some implementations, the current time-domain characteristic parameters include current kurtosis, current waveform factor, and current margin factor; the current frequency-domain characteristic parameters include the current spectrum integral of a preset frequency band; and the current time-frequency-domain characteristic parameters include the modulus maxima and variance of the wavelet coefficients of the sub-frequency bands after the three-level decomposition of the wavelet packet.

[0110] Among these, current kurtosis can be a statistical measure describing the shape of a current signal waveform, measuring the degree to which a data value deviates from the average. Current waveform factor can be a statistical parameter describing the shape of a current waveform, measuring the difference between the actual current waveform and an ideal waveform (such as a sine wave). Current margin factor can be a parameter measuring the fluctuation range or volatility of a current signal. Current spectrum integral can be a characteristic parameter describing the energy distribution of a current signal in the frequency domain. Wavelet packet three-level decomposition can be a method of decomposing a signal into different frequency components. Sub-bands can represent components of a signal within a specific frequency range. Wavelet coefficients can describe the changes in a signal at different frequencies and time scales.

[0111] Specifically, to avoid the problem that dimensional parameters are sensitive to current characteristics and easily affected by interference, dimensionless parameters such as current kurtosis, current waveform factor, and current margin factor are selected as current time-domain characteristic parameters. In the frequency domain, a comprehensive measure is needed to describe the frequency distribution of the current signal, while remaining independent of specific frequency values ​​and reducing the sensitivity of parameters to current changes. The current spectrum integral can be used as the current frequency-domain characteristic parameter. Furthermore, to reflect the characteristics of the current signal in both the time and frequency domains and to more accurately describe and analyze the time-frequency domain characteristics of the current signal, the modulus maxima and variance of the wavelet coefficients in the sub-bands after three-level wavelet packet decomposition can be used as current time-frequency domain characteristic parameters.

[0112] For example, the formulas for current kurtosis, current waveform factor, and current margin factor are obtained by extracting time-domain features from the current values ​​within the time window, as shown below:

[0113]

[0114] Where ku(s) is the current kurtosis, F(s) is the current waveform factor, and M(s) is the current margin factor. N is the number of sampling data points within the time window, and S... i S represents the amplitude of the sampled data points of the current value within the time window. max The maximum amplitude of the sampled data points of the current value within the time window. This represents the average amplitude of the sampled data points of the current value within the time window.

[0115] The frequency spectrum integral within the range of 30kHz to 125kHz after Fourier transform can be used as a characteristic parameter of the current frequency domain. The frequency spectrum integral is determined according to the following formula:

[0116]

[0117] Where C(x) is the spectrum integral, i.e. the current frequency domain characteristic parameter, N is the number of sampling data points in the time window, f1 represents the frequency band, f2 represents the frequency band, and x is the spectrum amplitude of the current value in the time window after Fourier transform. This can eliminate the influence of the number of sampling data points on the frequency domain characteristics of the current.

[0118] The current value within the time window can be decomposed into three levels of wavelet packets using the db4 wavelet (Daubechies 4th order), and the modulus maxima and variance of the wavelet coefficients in the sub-frequency band can be selected as the time-frequency domain features of the current.

[0119] In the above-mentioned DC series arc fault detection method, the current time-domain characteristic parameters include current kurtosis, current waveform factor and current margin factor, the current frequency-domain characteristic parameters include the current spectrum integral of the preset frequency band, and the current time-frequency domain characteristic parameters include the wavelet coefficient modulus maxima and variance of the sub-frequency band after the three-level decomposition of the wavelet packet. This provides a more comprehensive and accurate understanding of the current characteristics and improves the accuracy of DC series arc fault detection.

[0120] In some implementations, after performing principal component analysis to reduce the dimensionality of the current time-domain characteristic parameters, current frequency-domain characteristic parameters, and current time-frequency-domain characteristic parameters, the method may further include: normalizing the current characteristic parameters obtained by the dimensionality reduction process.

[0121] Specifically, in the current feature parameters obtained through dimensionality reduction, each current feature parameter may have different dimensions and numerical ranges. To eliminate these differences, the current feature parameters need to be normalized (e.g., standardized or scaled) to ensure that the current feature parameters have the same magnitude, thereby guaranteeing that the influence of each feature on the final result is equal during the training of the arc fault detection model.

[0122] For example, the formula for normalizing the current characteristic parameters obtained by dimensionality reduction is shown below:

[0123]

[0124] Among them, y i h is the result of normalizing the current characteristic parameters. i Let be the current characteristic parameters obtained by dimensionality reduction, min(h) be the minimum value among the current characteristic parameters obtained by dimensionality reduction, and max(h) be the maximum value among the current characteristic parameters obtained by dimensionality reduction.

[0125] In the above-mentioned DC series arc fault detection method, the current characteristic parameters obtained by dimensionality reduction are normalized to improve the performance of the arc fault detection model.

[0126] In some implementations, a voting mechanism is used to determine whether a DC series arc fault has occurred in the photovoltaic inverter based on multiple fault detection results. This may include: determining that a DC series arc fault has occurred in the photovoltaic inverter when at least two arc fault detection models have detected an arc fault in the photovoltaic inverter based on multiple fault detection results.

[0127] Specifically, a single arc fault detection model may misjudge or miss faults, while different arc fault detection models can capture the characteristics and manifestations of arc faults from different perspectives. Comprehensive analysis of the fault detection results from multiple arc fault detection models can enhance the reliability of detection, provide a more comprehensive understanding of the characteristics of arc faults, and thus improve the ability to diagnose and locate faults. Specifically, based on the fault detection results of each of the multiple arc fault detection models, it is determined whether each model detected an arc fault in the photovoltaic inverter. If at least two of the multiple arc fault detection models detect an arc fault in the photovoltaic inverter, it is determined that a DC series arc fault has occurred in the photovoltaic inverter.

[0128] In the above-mentioned DC series arc fault detection method, if at least two arc fault detection models detect an arc fault in the photovoltaic inverter based on multiple fault detection results, it is determined that the photovoltaic inverter has a DC series arc fault, thereby improving the reliability and accuracy of DC series arc fault detection.

[0129] This specification also provides a DC series arc fault detection method applied to a photovoltaic inverter. For an example, please refer to [link to example description]. Figure 7 The DC series arc fault detection method may include the following steps:

[0130] S702, Obtain the input voltage and input current of the photovoltaic inverter.

[0131] S704 performs voltage following and bandpass filtering on the input voltage, and calculates the processed voltage value based on the time window to obtain the voltage value within the time window.

[0132] S706. Extract time-domain and frequency-domain features from the voltage values ​​within the time window to obtain voltage time-domain feature parameters and voltage frequency-domain feature parameters.

[0133] Among them, the voltage time-domain characteristic parameters include voltage kurtosis, voltage waveform factor, and voltage margin factor, while the voltage frequency-domain characteristic parameters include the voltage spectrum integral of a preset frequency band.

[0134] S708: Filter and amplify the input current, and calculate the processed current value based on the time window to obtain the current value within the time window.

[0135] S710. Extract time-domain, frequency-domain, and time-frequency-domain features from the current value within the time window to obtain current time-domain feature parameters, current frequency-domain feature parameters, and current time-frequency-domain feature parameters.

[0136] Among them, the current time-domain characteristic parameters include current kurtosis, current waveform factor and current margin factor; the current frequency-domain characteristic parameters include the current spectrum integral of the preset frequency band; and the current time-frequency domain characteristic parameters include the modulus maxima and variance of the wavelet coefficients of the sub-frequency band after the three-level decomposition of the wavelet packet.

[0137] S712. Principal component analysis is used to reduce the dimensionality of the current time-domain characteristic parameters, current frequency-domain characteristic parameters, and current time-frequency-domain characteristic parameters.

[0138] S714. Normalize the voltage time-domain characteristic parameters and voltage frequency-domain characteristic parameters, and normalize the current characteristic parameters obtained by dimensionality reduction to obtain multiple sets of characteristic data.

[0139] S716. Input multiple sets of feature data into the pre-trained corresponding arc fault detection model for processing to obtain multiple fault detection results.

[0140] Each arc fault detection model is pre-trained based on a corresponding feature dataset. Each feature data training set is obtained by processing the input voltage and input current of the photovoltaic inverter under the conditions of an arc generator simulating a DC series arc fault in the photovoltaic inverter and the normal operation of the photovoltaic inverter.

[0141] S718. If, based on multiple fault detection results, at least two arc fault detection models detect an arc fault in the photovoltaic inverter, it is determined that a DC series arc fault has occurred in the photovoltaic inverter.

[0142] This specification provides a DC series arc fault detection device 800, applied to a photovoltaic inverter. Please refer to [link / reference]. Figure 8 The DC series arc fault detection device 800 includes: a processing module 810 and a fault detection module 820.

[0143] The processing module 810 is used to acquire the input side voltage and input side current of the photovoltaic inverter, and process the input side voltage and input side current respectively to obtain multiple sets of feature data, wherein each set of feature data includes voltage feature parameters and current feature parameters;

[0144] The fault detection module 820 is used to input the multiple sets of feature data into the corresponding pre-trained arc fault detection models for processing, obtain multiple fault detection results, and use a voting mechanism to determine whether the photovoltaic inverter has experienced a DC series arc fault based on the multiple fault detection results. Each arc fault detection model is pre-trained based on a corresponding feature dataset. Each feature data training set is obtained by processing the input voltage and input current of the photovoltaic inverter under the conditions of the arc generator simulating a DC series arc fault in the photovoltaic inverter and the normal operation of the photovoltaic inverter.

[0145] For a detailed description of the DC series arc fault detection device, please refer to the description of the DC series arc fault detection method above, which will not be repeated here.

[0146] This specification provides a photovoltaic inverter, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the above embodiments.

[0147] This specification provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the above embodiments.

[0148] One embodiment of this specification provides a computer program product including instructions that, when executed by a processor of a computer device, enable the computer device to perform the steps of the method described in any of the above embodiments.

[0149] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

Claims

1. A method for detecting DC series arc faults, characterized in that, Applied to photovoltaic inverters, the method includes: The input voltage and input current of the photovoltaic inverter are obtained, and the input voltage and input current are processed respectively to obtain multiple sets of feature data, wherein each set of feature data includes voltage feature parameters and current feature parameters; The multiple sets of feature data are respectively input into the corresponding pre-trained arc fault detection models for processing to obtain multiple fault detection results. Each arc fault detection model is pre-trained based on the corresponding feature dataset. Each feature data training set is obtained by processing the input voltage and input current of the photovoltaic inverter under the conditions of the arc generator simulating the DC series arc fault of the photovoltaic inverter and the normal operation of the photovoltaic inverter. Based on the multiple fault detection results, a voting mechanism is used to determine whether the photovoltaic inverter has experienced a DC series arc fault.

2. The method according to claim 1, characterized in that, Processing the input-side voltage includes: The input voltage is subjected to voltage following and bandpass filtering, and the processed voltage value is calculated based on a time window to obtain the voltage value within the time window; The voltage values ​​within the time window are subjected to time-domain and frequency-domain feature extraction to obtain voltage time-domain feature parameters and voltage frequency-domain feature parameters.

3. The method according to claim 2, characterized in that, The voltage time-domain characteristic parameters include voltage kurtosis, voltage waveform factor, and voltage margin factor, while the voltage frequency-domain characteristic parameters include the voltage spectrum integral of a preset frequency band.

4. The method according to claim 2, characterized in that, After obtaining the voltage time-domain characteristic parameters and voltage frequency-domain characteristic parameters, the method further includes: The voltage time-domain characteristic parameters and voltage frequency-domain characteristic parameters are normalized.

5. The method according to claim 1, characterized in that, Processing the input-side current includes: The input current is filtered and amplified, and the processed current value is calculated based on a time window to obtain the current value within the time window. The current value within the time window is subjected to time-domain features, frequency-domain features, and time-frequency-domain features to obtain current time-domain feature parameters, current frequency-domain feature parameters, and current time-frequency-domain feature parameters. Principal component analysis is used to reduce the dimensionality of the current time-domain characteristic parameters, the current frequency-domain characteristic parameters, and the current time-frequency-domain characteristic parameters.

6. The method according to claim 5, characterized in that, The current time-domain characteristic parameters include current kurtosis, current waveform factor, and current margin factor; the current frequency-domain characteristic parameters include the current spectrum integral of a preset frequency band; and the current time-frequency-domain characteristic parameters include the modulus maxima and variance of the wavelet coefficients of the sub-frequency bands after the three-level decomposition of the wavelet packet.

7. The method according to claim 5, characterized in that, After performing principal component analysis dimensionality reduction on the current time-domain characteristic parameters, the current frequency-domain characteristic parameters, and the current time-frequency-domain characteristic parameters, the method further includes: The current characteristic parameters obtained by dimensionality reduction are normalized.

8. The method according to any one of claims 1-7, characterized in that, Based on the multiple fault detection results, a voting mechanism is used to determine whether the photovoltaic inverter has experienced a DC series arc fault, including: If, based on the multiple fault detection results, at least two arc fault detection models detect an arc fault in the photovoltaic inverter, it is determined that the photovoltaic inverter has experienced a DC series arc fault.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the DC series arc fault detection method according to any one of claims 1-8.

10. A photovoltaic inverter, characterized in that, include: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the DC series arc fault detection method according to any one of claims 1-8.

11. A DC series arc fault detection device, characterized in that, The device, applied to photovoltaic inverters, includes: The processing module is used to acquire the input side voltage and input side current of the photovoltaic inverter, and process the input side voltage and input side current respectively to obtain multiple sets of feature data, wherein each set of feature data includes voltage feature parameters and current feature parameters; The fault detection module is used to input the multiple sets of feature data into the corresponding pre-trained arc fault detection models for processing, obtain multiple fault detection results, and use a voting mechanism to determine whether the photovoltaic inverter has experienced a DC series arc fault based on the multiple fault detection results. Each arc fault detection model is pre-trained based on a corresponding feature dataset. Each feature data training set is obtained by processing the input voltage and input current of the photovoltaic inverter under the conditions of an arc generator simulating a DC series arc fault in the photovoltaic inverter and under the condition of normal operation of the photovoltaic inverter.

Images