Direct current arc discharge detection method and device of photovoltaic inverter, medium and equipment

Abstract

The invention discloses a direct current arc discharge detection method and device of a photovoltaic inverter, a medium and equipment, and belongs to the technical field of arc detection. The method comprises the following steps: acquiring an alternating current component current of a direct current side current of the photovoltaic inverter by using a current sensor, converting the alternating current component current through a hardware circuit to obtain a direct current voltage signal, and sequentially performing signal amplification processing and band-pass filtering processing on the direct current voltage signal to obtain a processed signal of which switching noise is filtered; a microcontroller is used for sampling the processed signal after the switching noise is filtered out, fast Fourier transform is carried out on the sampled processed signal, and frequency spectrum data of the processed signal are obtained; calculating to obtain a corresponding power spectrum by using the frequency spectrum data of the processed signal; and processing the power spectrum by using the neural network model so as to judge whether the arc discharge phenomenon is generated on the direct current side of the photovoltaic inverter. According to the invention, high-precision and high-robustness identification of the direct-current arc under the complex working condition is realized.

Description

Technical Field

[0001] This application relates to the field of arc detection technology, and in particular to a DC arc detection method, apparatus, storage medium and electronic equipment for photovoltaic inverters. Background Technology

[0002] As the global energy structure shifts towards cleaner and lower-carbon energy, the demand for photovoltaic (PV) power generation, as a key renewable energy technology, is increasing worldwide. PV power generation systems typically generate electricity through PV arrays. These arrays contain numerous electrical connection points, such as joints and terminals. These connection points can generate electric arcs due to various reasons, including installation defects, material aging, or physical damage. Electric arcs pose a unique challenge to electrical safety.

[0003] Photovoltaic power generation systems primarily use direct current (DC), which can generate DC arcs. A DC arc is a sustained high-temperature plasma discharge formed when an air gap is broken down by voltage. DC arcs can directly burn out electrical equipment and carbonize insulation materials, and can also cause fires in photovoltaic power plants. Therefore, rapid and accurate detection and interruption of DC arcs are crucial in photovoltaic power generation systems to ensure the safety of assets and personnel.

[0004] To achieve rapid and accurate detection and interruption of DC arcs, arc fault detection technology has undergone several generations of evolution. Primary arc fault detection methods rely on monitoring whether the voltage or current of the circuit is abnormal. However, this method is almost ineffective for series arcs that do not cause significant voltage and current changes. Since photovoltaic power generation systems often generate a large number of series arcs, this method cannot effectively ensure the asset and personnel safety of photovoltaic power plants.

[0005] Subsequent arc fault detection methods analyze high-frequency signals in the circuit, as arc discharge generates unique broadband noise. Based on this characteristic, technicians use high-frequency current sensors (CTs) to capture circuit signals and process them using specialized computing units to determine if an arc has occurred. These computing units typically employ high-performance digital signal processors (DSPs), which execute digital signal processing algorithms such as Fast Fourier Transform (FFT) to analyze the energy distribution of the acquired signals within a specific frequency band, using this as a basis for determining the presence of an arc. However, limited by the performance of computing equipment and computational methods, this method still cannot quickly and accurately identify arcs, and therefore cannot guarantee the asset and personnel safety of photovoltaic power plants. Summary of the Invention

[0006] To address the problem that existing technologies cannot achieve rapid and accurate detection and interruption of DC arcs, this application mainly provides a DC arc detection method, device, storage medium, electronic equipment, and computer program product for photovoltaic inverters.

[0007] To achieve the above objectives, the first technical solution adopted in this application is: a DC arcing detection method for a photovoltaic inverter, comprising: acquiring the AC component current of the DC side current of the photovoltaic inverter using a current sensor, converting the AC component current into a DC voltage signal through hardware circuitry, and sequentially amplifying and bandpass filtering the DC voltage signal to obtain a processed signal with switched noise removed; sampling the processed signal using a microcontroller, and performing a fast Fourier transform on the sampled processed signal to obtain the spectrum data of the processed signal; calculating the corresponding power spectrum using the spectrum data of the processed signal; and processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

[0008] Optionally, the training process of the neural network model includes: acquiring first current data samples of the DC side of the photovoltaic inverter under normal operating conditions and second current data samples under arcing conditions; performing fast Fourier transform on the first current data samples and the second current data samples, and calculating the power spectrum based on the fast Fourier transform results to obtain the first current power spectrum and the second current power spectrum; using the first current power spectrum and the second current power spectrum to perform supervised training on the untrained neural network model to obtain the neural network model, so as to use the neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

[0009] Optionally, performing a Fast Fourier Transform on the sampled and processed signal to obtain the spectrum data of the processed signal, and processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter, respectively include: performing a Fast Fourier Transform using an embedded neural network processing unit, and performing a neural network model calculation using an embedded neural network processing unit to determine whether arcing occurs on the DC side of the photovoltaic inverter, wherein the embedded neural network processing unit is located within the microcontroller.

[0010] Optionally, processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter includes: inputting the power spectrum into the neural network model for processing, and the neural network model outputting the corresponding arcing probability, wherein the arcing probability reflects the possibility that the AC component current is the current generated by the arcing phenomenon; and determining whether arcing occurs on the DC side of the photovoltaic inverter based on the arcing probability and a preset arcing probability threshold.

[0011] Optionally, based on the arcing probability and a preset arcing probability threshold, it is determined whether an arcing phenomenon occurs on the DC side of the photovoltaic inverter, including: if the arcing probability is not less than the preset arcing probability threshold, then it is determined that an arcing phenomenon occurs on the DC side of the photovoltaic inverter; if the arcing probability is less than the preset arcing probability threshold, then it is determined that no arcing phenomenon occurs on the DC side of the photovoltaic inverter.

[0012] Optionally, if the arcing probability is not less than a preset arcing probability threshold, then determining that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter includes: within a predetermined time period, if the number of occurrences of the determination result that the arcing probability is not less than the preset arcing probability threshold is not less than a preset number threshold, then determining that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter.

[0013] Optionally, the first current data sample of the DC side of the photovoltaic inverter under normal operating conditions and the second current data sample under arcing conditions are obtained respectively, including: detecting the voltage difference between the two sides of the machine that generates the arcing phenomenon through a voltage detection circuit, determining whether the collected current data sample is the current generated at the moment of arcing; binding the current data sample with the corresponding tag according to the determination result, and binding the tag with the data synchronously sampled by the ADC of the microcontroller.

[0014] The second technical solution adopted in this application is: a DC arcing detection device for a photovoltaic inverter, comprising: a current acquisition module, used to acquire the AC component current of the DC side current of the photovoltaic inverter using a current sensor, and convert the AC component current into a DC voltage signal through hardware circuitry, and to sequentially amplify and bandpass filter the DC voltage signal to obtain a processed signal with switched noise removed; a spectrum data acquisition module, used to sample the processed signal with switched noise removed using a microcontroller, and to perform a fast Fourier transform on the sampled processed signal to obtain the spectrum data of the processed signal; a power spectrum acquisition module, used to calculate the corresponding power spectrum using the spectrum data of the processed signal; and a judgment module, used to process the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

[0015] The third technical solution adopted in this application is: a computer-readable storage medium storing a computer program / instruction, which is operated to execute the DC arcing detection method for photovoltaic inverters in Solution 1.

[0016] The fourth technical solution adopted in this application is: a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the DC arcing detection method for photovoltaic inverters in Solution 1.

[0017] The beneficial effects that the technical solution of this application can achieve are as follows: Through extensive data training and data augmentation, this application enables the neural network model to learn data under various working conditions, thereby intelligently distinguishing between real electric arc and non-electric arc data, maintaining high accuracy under various complex working conditions, and ultimately achieving high-precision, high-robustness and rapid identification of DC electric arc under complex working conditions. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of a specific embodiment of the DC arcing detection method for a photovoltaic inverter according to this application; Figure 2 This is an overall block diagram of the arcing detection system for the photovoltaic inverter in this application; Figure 3 This is an overall block diagram of the arc detection device of this application; Figure 4 This is a schematic diagram comparing the arc-drawn and non-arc-drawn data in the time domain and frequency domain of this application; Figure 5 This is a schematic diagram of an arc-drawing machine in the prior art; Figure 6 This is a schematic diagram of a specific embodiment of a DC arcing detection device for a photovoltaic inverter according to this application.

[0020] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0021] The preferred embodiments of this application will now be described in detail with reference to the accompanying drawings, so that the advantages and features of this application can be more easily understood by those skilled in the art, thereby providing a clearer and more definite definition of the scope of protection of this application.

[0022] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0023] The unique wide-spectrum noise generated during arc discharge is a key feature. In existing technologies, technicians typically use a high-performance digital signal processor (DSP) as the core of the computation, combined with an arc fault detection unit with a fixed discrimination threshold, to detect arc discharges. The specific implementation of this technology is as follows: First, a high-speed analog-to-digital converter (ADC) is connected to a DSP chip, continuously acquiring data from a current transformer on the DC side of the photovoltaic power station that monitors the current in real time. The DSP chip performs real-time FFT operations on the digital signal acquired by the current transformer, calculating the energy integral or spectral peak value of the acquired signal in the high-frequency range (e.g., 30 kHz to 100 kHz). Then, this calculation result is compared with a calibrated and static judgment benchmark value. If the calculated value exceeds the benchmark multiple times consecutively, an arc fault is considered to have occurred in the photovoltaic power station, triggering an alarm or protection action; if the calculated value does not exceed the benchmark multiple times consecutively, no arc fault is considered to have occurred in the photovoltaic power station, and no alarm or protection action is triggered.

[0024] Inverters (also called inverters or DC / AC converters, which are power conversion devices that convert DC power to AC power) are often used in the operation of photovoltaic systems. Inverters themselves generate switching noise during operation. Simultaneously, grid fluctuations in the photovoltaic system often introduce interference, and electromagnetic radiation from other electrical equipment in the environment, combined with these factors, create complex background noise in the circuit. The spectral characteristics of this noise sometimes highly overlap with those of a real electric arc, leading to "noise confusion." The aforementioned technical solutions rely on fixed discrimination thresholds to determine whether an arc has occurred. However, in complex environments, they cannot intelligently distinguish between background noise and arc signals, resulting in low accuracy in real-world scenarios and frequent false alarms and dangerous missed alarms.

[0025] Current technology compares the power spectrum corresponding to the current with a calibrated and static benchmark value. If the calculated value exceeds the benchmark multiple times consecutively, an arcing fault is considered to have occurred in the photovoltaic power station. This approach uses one or a few simple, linear rules to describe a complex, nonlinear physical phenomenon. However, the spectral characteristics of arcing and various noises are often intertwined and overlapping, making it impossible to clearly distinguish whether an arcing phenomenon has occurred using simple values ​​above or below a certain threshold. Therefore, current technology is only effective in controlled environments and struggles to adapt to changes in complex and dynamic real-world applications, resulting in unstable or low accuracy in the determination of arcing faults.

[0026] To avoid the inability to accurately and quickly detect electric arcs in complex environments, thus compromising the asset and personnel safety of photovoltaic power plants, this application constructs a complete electric arc detection solution based on a neural network model, encompassing data acquisition and actual deployment.

[0027] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with specific embodiments. The specific embodiments described below can be combined with each other to form new embodiments. The same or similar ideas or processes described in one embodiment may not be repeated in other embodiments.

[0028] Figure 1 This paper illustrates one embodiment of a DC arcing detection method for a photovoltaic inverter according to this application.

[0029] Figure 1 The DC arcing detection method for the photovoltaic inverter shown includes: Step S101, acquiring the AC component current of the DC side current of the photovoltaic inverter using a current sensor, converting the AC component current into a DC voltage signal using hardware circuitry, and then performing signal amplification and bandpass filtering on the DC voltage signal to obtain a processed signal with switched noise removed; Step S102, sampling the processed signal using a microcontroller, and performing a fast Fourier transform on the sampled processed signal to obtain the spectrum data of the processed signal; Step S103, calculating the corresponding power spectrum using the spectrum data of the processed signal; Step S104, processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

[0030] This specific implementation, through extensive data training and data augmentation, enables the neural network model to learn data under various working conditions, thereby intelligently detecting real electric arc and non-electric arc data. It distinguishes variable background noise as non-electric arc data, thus enabling the application to maintain high accuracy under various complex working conditions. Ultimately, it achieves high-precision, high-robustness, and rapid identification of DC electric arcs under complex working conditions.

[0031] Specifically, this application employs a low-cost microcontroller (MCU) with an embedded neural network processing unit (NPU) as the computing core, and loads both the Fast Fourier Transform (FFT) and neural network inference—two computationally intensive tasks—onto the NPU hardware for execution. This significantly reduces the computational burden on the CPU within the MCU, achieving a response speed that surpasses traditional DSP solutions. Furthermore, since training the neural network model requires a large amount of labeled data, this invention combines hardware circuitry and software programs, utilizing methods such as... Figure 2 The structure shown can automatically determine the presence of arcing phenomena during training data collection and assign corresponding labels to the data, solving the problem of encountering incorrect data when manually labeling datasets. Through extensive data training and data augmentation, the neural network model can learn data under various operating conditions, thereby intelligently detecting real electric arcs and maintaining high accuracy under various complex operating conditions. Ultimately, it achieves high-precision, high-robustness, and rapid identification of DC arcs under complex operating conditions.

[0032] In one specific embodiment of this application, the DC arcing detection method for a photovoltaic inverter includes step S101, which involves using a current sensor to acquire the AC component current of the DC side current of the photovoltaic inverter, converting the AC component current into a DC voltage signal through hardware circuitry, and then performing signal amplification and bandpass filtering on the DC voltage signal in sequence to obtain a processed signal with switching noise removed.

[0033] Specifically, such as Figure 3 As shown, the AC component of the DC-side current of the photovoltaic inverter is acquired by a current sensor, and converted into a DC voltage signal by a sampling resistor and a 1.65V bias voltage. This voltage signal is then amplified by an amplifier circuit. The amplified signals are then passed through a bandpass filter consisting of a 100kHz low-pass filter and a 20kHz high-pass filter to remove switching noise and obtain a processed signal. Finally, the filtered and processed signal is acquired by an ADC. The parameters of the bandpass filter here are for illustrative purposes only.

[0034] In one specific embodiment of this application, the DC arcing detection method for a photovoltaic inverter includes step S102, which involves sampling the processed signal using a microcontroller and performing a fast Fourier transform on the sampled signal to obtain the corresponding spectrum data, and step S103, which involves calculating the corresponding power spectrum using the spectrum data. Specifically, the filtered signal is sampled from the hardware circuit by the ADC of the main control chip MCU. The sampling rate can be 250kHz and the number of sampling points can be 2048. Then, the sampled and processed signal is subjected to a fast Fourier transform to obtain the spectrum data of the processed signal. Finally, the corresponding power spectrum is calculated using the spectrum data of the processed signal.

[0035] In one specific embodiment of this application, the DC arcing detection method for a photovoltaic inverter includes step S104, which uses a neural network model to process the power spectrum to determine whether an arcing phenomenon occurs on the DC side of the photovoltaic inverter.

[0036] Specifically, the power spectrum is input into a pre-learned neural network model. The neural network model can automatically identify the presence of local morphological features representing an electric arc within the power spectrum, and thus determine in the microprocessor whether arcing has occurred on the DC side of the photovoltaic inverter. The neural network model in this application is primarily a one-dimensional convolutional neural network (CNN). This application utilizes an NPU for inference calculations of the CNN neural network model, accelerating the computational inference process and enabling faster output of the arcing probability of the current data.

[0037] In this application, the last layer of the CNN neural network model that has completed learning is a 1x1 convolutional layer. This convolutional layer is responsible for weighted summation of the feature values ​​of all previous convolutional layers and mapping all the features it receives into a single value, thereby representing the probability that the AC current component corresponding to the power spectrum in the input neural network model is generated by arcing. The single value output here is called the original score logit, which represents the original confidence score of the neural network model in judging that the AC current component corresponding to the current input is "arcing".

[0038] Preferably, the maximum value of the original score is 1. In order to convert the logit value, which can be of any size, into a standard probability between 0 and 1, this application applies the Sigmoid activation function at the end of the model. The Sigmoid function is used to map the logit value output by the 1x1 convolutional layer. The output of the CNN neural network model can directly and intuitively reflect the probability of arc generation.

[0039] In one specific embodiment of this application, processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter includes: inputting the power spectrum into the neural network model for processing, the neural network model outputting the corresponding arcing probability, wherein the arcing probability reflects the probability that the AC component current on the DC side of the photovoltaic inverter is the current generated by the arcing phenomenon; and determining whether arcing occurs on the DC side of the photovoltaic inverter based on the arcing probability and a preset arcing probability threshold.

[0040] In one specific embodiment of this application, determining whether an arcing phenomenon occurs on the DC side of the photovoltaic inverter based on the arcing probability and a preset arcing probability threshold includes: if the arcing probability is not less than the preset arcing probability threshold, then determining that an arcing phenomenon occurs on the DC side of the photovoltaic inverter; if the arcing probability is less than the preset arcing probability threshold, then determining that no arcing phenomenon occurs on the DC side of the photovoltaic inverter.

[0041] Specifically, when the arcing probability output by the neural network model is not less than a preset arcing probability threshold, it is used to determine whether arcing has occurred on the DC side of the photovoltaic inverter. The arcing probability threshold can be set to 0.90. That is, when the arcing probability output by the neural network model is not less than 0.90, it is considered that arcing exists in the current data sampled by the microprocessor. A higher preset arcing probability threshold results in fewer false alarms but potentially more missed alarms. In practical applications, for arcing detection in photovoltaic power plants, a lower false alarm rate is preferred. Therefore, this application sets a higher arcing probability threshold in its example. In actual applications, developers can adjust the specific value of the arcing probability threshold according to actual needs.

[0042] In one specific embodiment of this application, if the arcing probability is not less than a preset arcing probability threshold, then determining that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter includes: within a predetermined time period, if the number of occurrences of the determination result that the arcing probability is not less than the preset arcing probability threshold is not less than a preset number threshold, then determining that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter.

[0043] Specifically, in order to further reduce the false alarm rate, this application counts the number of times the arcing probability output by the neural network model is not less than a preset arcing probability threshold within a predetermined time period. If the number of occurrences is not less than the preset threshold, it is determined that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter.

[0044] For example, during 10 consecutive tests, the number of times the arcing probability is not less than the arcing probability threshold is counted. If this number is not less than 7, then the DC side of the photovoltaic inverter is considered to have experienced arcing. That is, if there are more than 7 events with an arcing probability greater than 0.90 in 10 consecutive inference results, then the arcing phenomenon is considered to have indeed occurred. In this case, the main control chip will report the information to the cloud through information transmission modules such as 4G modules and activate alarms and other related protection measures.

[0045] Based on the data from the above example, the current data acquisition time of a microprocessor's ADC (analog-to-digital converter) is 8ms, and the total time for ADC data processing and model inference is about 3ms. Since the ADC acquisition information in this application is a hardware-executed action, the NPU inference operation of the previous set of data can be performed while the ADC is acquiring the next set of data. Therefore, the time from the start of data acquisition to obtaining the result is 11ms, and for 10 sets of data, it is 83ms. This is still a fairly short time. Even with the above-mentioned technical means to determine whether arcing has occurred, this application can still quickly determine whether arcing has occurred.

[0046] In one specific embodiment of this application, a fast Fourier transform is performed using an embedded neural network processing unit, and a neural network model is used to calculate whether arcing occurs on the DC side of the photovoltaic inverter. The embedded neural network processing unit is located within the microcontroller.

[0047] Specifically, by accelerating both the Fast Fourier Transform (FFT) and model inference through parallel computing using high-efficiency NPU (embedded neural network processing unit) hardware, the data processing link latency for signal acquisition, feature extraction, and intelligent judgment in this application is extremely low. Therefore, this application can achieve instantaneous response to electric arcs, quickly identifying and outputting alarms in the early stages of an electric arc (milliseconds), greatly shortening the time from the occurrence of a fault to the system response, and gaining valuable time to cut off the circuit and prevent fires, thereby effectively improving the overall safety of the photovoltaic system.

[0048] This application accelerates the deployment of CNN models on an NPU, not by simply running them directly, but by a combination of dedicated NPU hardware design and a series of software optimization processes. Since CPUs are general-purpose processors designed to handle diverse tasks, their computing units occupy a relatively small proportion of the chip area. In contrast, NPUs are processors specifically designed for neural network inference computation, integrating hundreds or thousands of tiny multiply-accumulate (MAC) computation units. Most computations in CNN models involve convolution, which is essentially matrix multiplication. Matrix multiplication can be decomposed into individual multiply-accumulate operations. Therefore, the NPU can utilize a large number of MAC computation units to perform these multiply-accumulate operations in parallel. However, if the CPU were used for these computations, it would require multiple iterations of the multiply-accumulate process, resulting in a serial computation to obtain the final result. Therefore, the computational efficiency of using an NPU for neural network model inference in this application is far higher than that of using a CPU. Furthermore, to enable the NPU to infer models more efficiently, this application also improves computational efficiency through a series of software optimization processes, such as model quantization, operator fusion, and graph optimization. After these software processing steps, the original CNN model is compiled into a highly optimized instruction set that the NPU can directly load and execute. Therefore, the CPU in the MCU only needs to transmit the ADC data to be processed to the NPU, and the NPU can perform FFT calculations and neural network model inference calculations in a very short time, thereby achieving the effect of accelerating inference.

[0049] Among them, the model quantization process is the key to enabling neural network models to infer faster. Quantization mainly converts the 32-bit floating-point (FP32) neural network model into an 8-bit fixed-point integer (INT8) model that the NPU is better at processing. This makes the neural network model faster in a single operation, with less memory usage and lower power consumption.

[0050] During operator fusion and graph optimization computation, the software toolchain automatically analyzes the computation process of the neural network model. Then, it merges multiple consecutive operations in the computation process (such as batch normalization after convolution computation and activation function processing) into a single hardware instruction, reducing the overhead of data transfer and further improving inference speed.

[0051] Meanwhile, the NPU also contains a Parallel Processing Unit (PPU) that supports Single Instruction Multiple Data (SIMD). This function is mainly used to process parts that the MAC cannot directly calculate, such as exponential, logarithmic, and floating-point calculations. Therefore, for the FFT calculation and power spectrum calculation of the ADC, the PPU can process multiple FP32 data at once. This application utilizes this feature of the NPU to accelerate the FFT calculation.

[0052] In one specific embodiment of this application, the training process of the neural network model includes: acquiring first current data samples of the DC side of the photovoltaic inverter under normal operating conditions and second current data samples under arcing conditions; performing fast Fourier transform on the first and second current data samples, and calculating the power spectrum based on the fast Fourier transform results to obtain the first current power spectrum and the second current power spectrum; using the first and second current power spectra to perform supervised training on the neural network model to obtain a trained neural network model, which is then used to determine whether arcing occurs on the DC side of the photovoltaic inverter. By using a CNN neural network model trained with deep learning to determine whether arcing has occurred, this CNN model can learn and recognize the complex and nonlinear characteristics of the arc signal on the power spectrum, thereby more accurately detecting real arcs. This allows the application to maintain high accuracy under various complex operating conditions, fundamentally solving the technical problem that traditional methods are prone to false alarms or missed alarms due to environmental changes.

[0053] Specifically, the CNN model training process used in this application avoids the need for manual design and extraction of complex features in traditional methods. The CNN model in this application learns directly from the original frequency domain data and then distinguishes the probability of arcing.

[0054] The specific learning process of the CNN model used in this application is as follows: 1. Dataset Establishment: First, this application collects a large amount of data from current transformers through the ADC of the MCU, which is then amplified and bandpass filtered. This data includes positive sample current data and negative sample current data of the inverter under various operating conditions. The sample data is labeled with corresponding tags by the voltage difference detection circuit on both sides of the arcing machine. The tag indicates whether arcing has occurred. Positive sample current data refers to the data when the inverter generates arcing, and negative sample data refers to the data when arcing has not occurred.

[0055] 2. Input Data Construction: Each consecutive sample of 2048 ADC data points and their corresponding labels is considered a complete training data sample. These 2048 ADC data points are first processed using a Fast Fourier Transform (FFT) to calculate their corresponding power spectra. The length of the power spectrum can be 1024, and the frequency range can be 0Hz to 125kHz. These 1024 power spectrum data points are the input data for training the CNN model, and the positive and negative sample labels corresponding to the sample data are the true labels needed for training the CNN model.

[0056] 3. Model Training: The labeled power spectrum data is input into the CNN model for supervised learning. During training, the CNN's convolutional kernels slide across the input one-dimensional power spectrum vector. Each kernel can be considered a feature detector. Initially, the data in these kernels is random, but through backpropagation and by continuously adjusting the weights of these kernels using an optimizer, the data gradually becomes more regular. The optimizer can be Adam. The goal of training is to minimize the loss between the model's output (the predicted probability of arcing) and the true label (1 or 0). The training process iterates continuously. Through comparison with a large number of positive and negative samples, the convolutional kernels in the model can extract the most effective frequency domain morphological features to distinguish between "arcing" and "non-arcing." Furthermore, these features are often more complex and robust than rules manually defined by human experts.

[0057] Figure 4 This is a schematic diagram comparing the arc-drawn and non-arc-drawn data in the time domain and frequency domain of this application. Figure 4 For example, in the time domain, there is no significant difference between the arc-drawn and non-arc-drawn waveforms. However, in the frequency domain, through comparison... Figure 4 In the mid-range arc waveform, there is an energy rise in the 20kHz~60kHz range, and several sharp spectral peaks with fixed positions exist. These peaks represent noise generated by the inverter's own switching devices, or noise generated by some electrical loads. And through... Figure 4 It can be seen that, apart from the fixed switching frequency noise, the characteristics of arcing are difficult to distinguish in a formulaic way, not to mention that there are other subtle differences in the arcing spectrum, which are even more difficult to represent and distinguish. However, the CNN neural network model can automatically learn these data characteristics of arcing and non-arcing through the above process, and can distinguish the inherent noise and load noise in the circuit. It does not require manual definition of relevant rules, making it more practical and convenient.

[0058] Furthermore, the method acquires first current data samples from the DC side of the photovoltaic inverter under normal operating conditions and second current data samples under arcing conditions. This includes: detecting the voltage difference between the two sides of the machine controlling the arcing phenomenon using a voltage detection circuit to determine whether the collected current data sample is the current generated at the moment of arcing; binding a corresponding tag to the current data sample based on the determination result, and binding the tag to the data synchronously sampled by the microcontroller's ADC. This application, by using a voltage detection circuit to detect the existence of a voltage difference during the data acquisition stage, thereby determining the arcing state of the collected current data sample in real time and generating a hardware-level "tag," then binding the tag to the synchronously acquired ADC data, and finally uploading it to the host computer via a USB interface, allows the sample data needed for training a neural network model to be obtained without the need for manual labeling of sample data attributes by sampling personnel. This solves the problem of obtaining high-quality, labeled sample data for neural network models.

[0059] Specifically, using, such as Figure 5 The arc-generating machine shown in this application is used in a laboratory environment to acquire data by controlling whether arcing occurs. The point of arcing is known (i.e., it occurs on the arc-generating machine), and a voltage difference is generated across the arc-generating machine when arcing occurs. Based on this, the comparator in the voltage detection circuit can detect whether the voltage across the arc-generating machine is within a certain range, thereby determining whether the acquired current data sample is the current generated by the arcing phenomenon. Simultaneously, the high and low levels output by the comparator in the voltage detection circuit transmit information to the main control chip regarding whether the currently sampled data is the data at the moment of arcing. Afterwards, the main control chip combines its ADC sampled data and arcing tag data and sends it to the host computer in real time via USB interface. The host computer software receives and saves this data for subsequent training of the neural network model.

[0060] Figure 6 This paper illustrates a specific embodiment of a DC arcing detection device for a photovoltaic inverter according to this application.

[0061] exist Figure 6 In the specific embodiment shown, the DC arcing detection device of the photovoltaic inverter mainly includes: a current acquisition module 601, which is used to acquire the AC component current of the DC side current of the photovoltaic inverter using a current sensor, and convert the AC component current into a DC voltage signal using hardware circuitry, and then perform signal amplification and bandpass filtering on the DC voltage signal in sequence to obtain a processed signal with switching noise removed. The spectrum data acquisition module 602 is used to sample the processed signal using a microcontroller and perform a fast Fourier transform on the sampled processed signal to obtain the spectrum data of the DC voltage signal. The power spectrum acquisition module 603 is used to calculate the corresponding power spectrum using the spectrum data of the processed signal; the judgment module is used to process the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

[0062] The DC arcing detection device for photovoltaic inverters provided in this application can be used to perform the DC arcing detection method for photovoltaic inverters described in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0063] In one specific embodiment of this application, the functional modules of the DC arcing detection device for a photovoltaic inverter can be directly in hardware, in software modules executed by a processor, or in a combination of both.

[0064] Software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in this art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium.

[0065] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor can be a microprocessor, but alternatively, it can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors incorporating a DSP core, or any other such configuration. Alternatively, the storage medium can be integrated with the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in the user terminal. Alternatively, the processor and storage medium can reside as discrete components in the user terminal.

[0066] In another specific embodiment of this application, a computer-readable storage medium is provided, which stores a computer program / instructions that are operated to perform the DC arcing detection method for a photovoltaic inverter described in the above embodiments.

[0067] In one specific embodiment of this application, a computer device includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the DC arcing detection method for a photovoltaic inverter described in the above embodiments.

[0068] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0069] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0070] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for detecting DC arcing in a photovoltaic inverter, characterized in that, include: The AC component of the DC side current of the photovoltaic inverter is obtained by using a current sensor, and the AC component current is converted into a DC voltage signal by hardware circuit. The DC voltage signal is then amplified and bandpass filtered in sequence to obtain a processed signal with switching noise removed. The processed signal is sampled using a microcontroller, and the sampled processed signal is subjected to a fast Fourier transform to obtain the spectrum data of the processed signal; The corresponding power spectrum is calculated using the spectral data of the processed signal; The power spectrum is processed using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

2. The DC arcing detection method for photovoltaic inverters according to claim 1, characterized in that, The training process of the neural network model includes: The first current data sample of the DC side of the photovoltaic inverter under normal operating conditions and the second current data sample under arcing conditions are obtained respectively. Perform a fast Fourier transform on the first current data sample and the second current data sample, and calculate the power spectrum based on the result of the fast Fourier transform to obtain the first current power spectrum and the second current power spectrum. The untrained neural network model is supervised and trained using the first current power spectrum and the second current power spectrum to obtain the neural network model, which is then used to determine whether arcing occurs on the DC side of the photovoltaic inverter.

3. The DC arcing detection method for photovoltaic inverters according to claim 1, characterized in that, Performing a Fast Fourier Transform on the sampled and processed signal to obtain the spectrum data of the processed signal, and processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter, respectively include: The embedded neural network processing unit is used to perform the Fast Fourier Transform and to perform the calculation of whether arcing occurs on the DC side of the photovoltaic inverter using the neural network model. The embedded neural network processing unit is located within the microcontroller.

4. The DC arcing detection method for a photovoltaic inverter according to claim 1, characterized in that, The step of processing the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter includes: The power spectrum is input into the neural network model for processing, and the neural network model outputs the corresponding arcing probability, wherein the arcing probability reflects the possibility that the AC component current is an arcing phenomenon. Based on the arcing probability and the preset arcing probability threshold, it is determined whether arcing occurs on the DC side of the photovoltaic inverter.

5. The DC arcing detection method for a photovoltaic inverter according to claim 4, characterized in that, The step of determining whether arcing occurs on the DC side of the photovoltaic inverter based on the arcing probability and a preset arcing probability threshold includes: If the arcing probability is not less than the preset arcing probability threshold, then it is determined that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter. If the arcing probability is less than the preset arcing probability threshold, it is determined that no arcing phenomenon has occurred on the DC side of the photovoltaic inverter.

6. The DC arcing detection method for a photovoltaic inverter according to claim 5, characterized in that, If the arcing probability is not less than the preset arcing probability threshold, then determining that arcing has occurred on the DC side of the photovoltaic inverter includes: If, within a predetermined time period, the number of occurrences of the judgment result that the arcing probability is not less than a preset arcing probability threshold is not less than a preset number threshold, then it is determined that an arcing phenomenon has occurred on the DC side of the photovoltaic inverter.

7. The DC arcing detection method for a photovoltaic inverter according to claim 2, characterized in that, The process of acquiring the first current data sample of the DC side of the photovoltaic inverter under normal operating conditions and the second current data sample under arcing conditions includes: The voltage detection circuit detects the voltage difference between the two sides of the machine that generates the arcing phenomenon, and determines whether the collected current data sample is the current generated at the moment of arcing. Based on the judgment result, the current data sample is bound to the corresponding tag, and the tag is bound to the data synchronously sampled by the ADC of the microcontroller.

8. A DC arcing detection device for a photovoltaic inverter, characterized in that, include: The current acquisition module is used to acquire the AC component current of the DC side current of the photovoltaic inverter using a current sensor, convert the AC component current into a DC voltage signal through hardware circuitry, and perform signal amplification and bandpass filtering on the DC voltage signal in sequence to obtain a processed signal with switching noise removed. The spectrum data acquisition module is used to sample the processed signal using a microcontroller and perform a fast Fourier transform on the sampled processed signal to obtain the spectrum data of the processed signal. The power spectrum acquisition module is used to calculate the corresponding power spectrum using the spectral data of the processed signal; The judgment module is used to process the power spectrum using a neural network model to determine whether arcing occurs on the DC side of the photovoltaic inverter.

9. A computer-readable storage medium storing a computer program / instructions, characterized in that, The computer program / instructions are operated to perform the DC arcing detection method for a photovoltaic inverter as described in any one of claims 1-7.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the DC arcing detection method for a photovoltaic inverter as described in any one of claims 1-7.

Images