Photovoltaic array direct current arc fault detection method and system based on group string current difference characteristics

Abstract

The application provides a photovoltaic array direct current arc fault detection method and system based on group string current difference characteristics. The method comprises the following steps: S1. Multi-channel synchronous current signal acquisition; S2. Layered signal preprocessing and data correction; S3. Full-channel signal FFT frequency domain conversion and feature frequency band demarcation; S4. Single-path group string feature frequency band energy accurate calculation; S5. Group string reference energy solving and abnormal data filtering; S6. Group string relative difference rate modeling calculation; S7. Multi-level steady-state threshold fault discrimination. The application relies on the same origin characteristics of group string working conditions under the same bus box, and takes the relative difference between group strings as the core fault basis, rather than the absolute amplitude of the single signal. Global interference acts on all group strings, and there is no obvious difference between group strings, which is directly filtered. Only when a single branch arc fault occurs, a significant difference will be produced, which distinguishes interference from fault from the root, effectively filters various conventional interferences, and greatly improves the detection accuracy.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic system safety monitoring and DC arc fault detection technology, specifically relating to a method and system for detecting DC arc faults in photovoltaic arrays based on string current difference characteristics. Background Technology

[0002] With the large-scale construction of distributed photovoltaic (PV) and centralized ground-mounted PV power plants, PV arrays are exposed to complex outdoor conditions for extended periods. Problems such as cable aging, loose joints, external wear, and microcracks in modules frequently occur, making them highly susceptible to DC series arcing faults. DC arcing is characterized by low arc voltage, difficulty in extinguishing, and strong high-temperature ignition, making it a core safety hazard causing PV power plant fires, equipment damage, and system shutdowns. Therefore, DC arcing fault detection has become a core research direction for the safe operation and maintenance of PV systems.

[0003] Currently, several mature technical solutions have been developed both domestically and internationally for photovoltaic (PV) DC arc faults. The mainstream technologies mainly include single-current frequency domain analysis, voltage characteristic detection, and partial discharge / acoustic-optical-assisted detection. Among these, voltage characteristic detection relies on the DC-side voltage fluctuation characteristics of the PV array for fault identification. This method identifies faults by monitoring voltage distortion, drops, and ripple changes at both ends of the string. However, PV arrays are affected by light intensity, ambient temperature, and cloud cover, resulting in large dynamic voltage fluctuations, making it only suitable for scenarios with severe arc faults and significant voltage changes. Partial discharge and acoustic-optical-assisted detection technologies use microphones, acoustic sensors, and photoelectric sensors to collect the sound and light signals of the arc for fault location. These methods are susceptible to interference from outdoor wind noise, equipment noise, and changes in light intensity, making them only suitable for localized scenarios such as indoor distribution cabinets and enclosed chambers, and unsuitable for widespread use in large-area outdoor PV arrays.

[0004] Single-current frequency domain analysis is currently the most widely used detection method. Its principle is as follows: a single-channel DC current signal from the photovoltaic string is acquired, and the time-domain current is converted into a frequency-domain signal using Fast Fourier Transform (FFT). High-frequency harmonics and stray spectrum characteristics generated by the arc are extracted, and a fixed threshold is set to determine whether an arc fault has occurred. This technology has a simple hardware architecture and easy-to-implement algorithms, and is widely integrated into combiner boxes and inverter-related monitoring devices. In practical engineering applications, DC series arc faults are the most hazardous and difficult-to-detect type of fault in photovoltaic arrays. Series arcs occur in series circuits such as string cables, terminals, and combiner buses. The fault current fluctuation amplitude is small, and the time-domain characteristics are weak. Traditional single-frequency domain detection schemes have become the mainstream choice for outdoor photovoltaic arrays and are currently the most common technical method in the industry. However, existing current frequency domain analysis techniques also have the following drawbacks in specific applications:

[0005] (1) Traditional frequency domain analysis methods have weak anti-interference capabilities, resulting in high false alarm and false negative rates. During the operation of photovoltaic inverters, switching frequency harmonics and stray noise of 20kHz~100kHz are generated. The spectral distribution of such system noise highly overlaps with the high-frequency characteristic band of DC arc faults, which will seriously drown out the weak arc fault characteristic signals. At the same time, the disturbances of normal operating conditions such as sudden changes in photovoltaic power station illumination, component temperature drift, and load switching will cause frequency domain energy fluctuations. Existing methods based on single string frequency domain characteristic thresholds cannot distinguish between normal operating disturbances, system noise interference, and real arc faults, resulting in a large number of false alarms and false negatives. According to industry test data, the false alarm rate of traditional frequency domain detection methods under dynamic operating conditions can reach more than 15%.

[0006] (2) Single-channel signal analysis mode has a detection blind zone and cannot adapt to the different operating conditions of strings; all existing current frequency domain analysis detection methods analyze the current signal of a single photovoltaic string independently, without utilizing the physical characteristics of multiple strings operating under the same combiner box and synchronous fluctuations; all photovoltaic strings under the same combiner box have basically the same installation environment, light conditions, and temperature conditions, and the current fluctuations of each string are highly synchronous during normal operation; however, arc faults are local faults of a single string, and only the faulty string will show current distortion and frequency domain energy abnormalities, while the non-faulty strings remain stable. Traditional single-channel detection mode cannot utilize this differentiated feature, making it difficult to eliminate global common interference, and the accuracy of fault identification is limited.

[0007] (3) Existing optimization algorithms are costly and have poor implementation. Although optimization schemes such as wavelet transform and artificial intelligence recognition can improve anti-interference ability to a certain extent, the algorithms are complex and consume a lot of computing power. They require high-end processors and cannot be adapted to low-cost, low-computing-power hardware devices such as photovoltaic combiner boxes and edge acquisition terminals. They are difficult to be widely used. The industry urgently needs a DC arc detection scheme with low computing power, high accuracy and strong anti-interference.

[0008] This invention addresses the core shortcomings of existing technologies by abandoning the traditional single-channel independent analysis detection logic. Instead, it conducts fault detection based on the current difference characteristics of multiple strings in the same combiner box. This aims to solve the industry technical problems of traditional frequency domain analysis methods, which are easily affected by inverter switching noise and environmental conditions, resulting in high false alarm rates, low detection accuracy, and poor hardware compatibility. Summary of the Invention

[0009] This invention addresses the shortcomings of existing photovoltaic (PV) DC series arc fault detection technologies, which rely on single-string frequency domain characteristics, have weak common-mode interference resistance, high false alarm rates for weak faults, and frequent false alarms. It innovatively proposes a PV array DC arc fault detection method and system based on string current difference characteristics. This method abandons the inherent logic of traditional independent detection of a single branch. Relying on the operating characteristics of PV strings operating under the same combiner box and experiencing similar interference sources, it constructs a full-link detection system consisting of "multi-channel synchronous data acquisition - precise signal preprocessing - frequency domain feature energy quantization - string difference feature modeling - multi-level steady-state fault discrimination." This system isolates common-mode interference from inverter switching frequency interference, global environmental electromagnetic noise, and light and temperature fluctuations from a technical mechanism perspective. Simultaneously, it amplifies the heterogeneous characteristics of single-branch arc faults, achieving high-precision, low-false-alarm detection and accurate location of early, weak, and steady-state PV DC series arc faults.

[0010] To achieve the above technical objectives, this invention provides a method for detecting DC arc faults in photovoltaic arrays based on string current difference characteristics, specifically including the following steps:

[0011] S1. Multi-channel fully synchronous current signal acquisition; taking a single combiner box as an independent detection unit, perform full-channel synchronous data acquisition on all N photovoltaic strings connected to it, continuously acquire the raw time domain data of DC current of each string, form N-channel raw current datasets with consistent dimensions and time sequence alignment, and synchronously transmit them to the preprocessing module, where N≥2;

[0012] S2. Layered signal preprocessing and data correction: After layered signal preprocessing of each raw current data collected in step S1, the preprocessing module outputs a standardized time-domain current signal with a significantly improved signal-to-noise ratio, and ensures that the processing algorithms and parameters of each signal are completely consistent, so as to ensure that the data has horizontal comparability.

[0013] S3. Full-channel signal FFT frequency domain conversion and characteristic frequency band delineation: For each pre-processed time-domain current sequence, perform fast Fourier transform one by one to completely convert the discrete time-domain current signal into a discrete frequency-domain amplitude signal, and obtain continuous frequency domain spectrum distribution data from 0Hz to 500Hz.

[0014] S4. Precise calculation of energy in the characteristic frequency band of a single string; For the frequency domain data of the current of each string, the total energy of the fault characteristic frequency band of a single string is calculated in the characteristic frequency band of 50Hz~300Hz through the spectrum energy integration algorithm, and the characteristic energy dataset of N strings under the current sampling period is generated.

[0015] S5. Solving the baseline energy of the string and filtering out abnormal data; preprocessing and filtering the feature energy dataset generated in step S4 to remove instantaneous extreme values ​​that deviate from the mean by more than 50% within a single period; then calculating the arithmetic mean of the energy values ​​of the remaining valid normal strings, which is defined as the global baseline energy Eavg under the current operating conditions of the combiner box;

[0016] S6. Modeling and Calculation of Relative Difference Rate of Strings; Constructing a relative difference rate calculation model of strings as the core evaluation index, and quantifying the degree of feature deviation of a single string by calculating the relative difference rate of each string to achieve the essential distinction between interference and fault;

[0017] S7. Multi-level steady-state threshold fault discrimination: A discrimination mechanism of two-level threshold + multi-cycle steady-state verification is adopted. When the relative difference rate of the string group is continuously greater than the fault diagnosis threshold and is maintained for 3 or more consecutive sampling cycles, it is formally determined that the string group has a DC series arc fault.

[0018] A further technical solution of the present invention: The method further includes: S8. Fault location and hierarchical output; After the fault is diagnosed, the system automatically locks the unique number of the abnormal string and synchronously executes hierarchical output: Locally triggers audible and visual fault alarms to remind on-site operation and maintenance personnel; Remotely uploads the fault type, fault string location, fault occurrence time, and fault characteristic data to the photovoltaic power station operation and maintenance platform through 4G / Ethernet communication to achieve accurate fault location and real-time source tracing, while continuously monitoring the fault status until the fault is eliminated and the data returns to normal.

[0019] A further technical solution of the present invention: In step S1, all sampling channels have consistent sampling accuracy and sampling response speed, and the sampling parameters are uniformly configured: the sampling frequency is fixed at 1kHz or above, the single-frame sampling time window is strictly locked at 1s, and the number of sampling points per channel per time is not less than 1000 to ensure the integrity of the time domain signal; all sampling channels share the same GPS or local high-precision synchronization clock, and the sampling timing error between channels is ≤1μs, completely eliminating the phase deviation and data misalignment problems caused by asynchronous sampling.

[0020] A further technical solution of the present invention: The S2 step of layered preprocessing specifically includes: First, using a first-order Butterworth low-pass filter algorithm to filter out irregular noise above 10kHz, specifically including ultra-high frequency instantaneous spike pulses, electrostatic interference, and instantaneous electromagnetic pulses from equipment; Second, performing baseline calibration processing to eliminate the current amplitude baseline offset problem caused by sampling hardware temperature drift and line voltage drop; Third, removing single-point abnormal extreme values ​​in a single frame of data and retaining continuous and stable effective current waveforms.

[0021] The preferred technical solution of the present invention is: in step S3, for each channel of preprocessed 1s time-domain current sequence.

[0022] A further technical solution of the present invention: The specific calculation method in step S4 is as follows: traverse the amplitude parameters of all frequency points in the characteristic frequency band, accumulate and integrate the squares of the amplitudes of each frequency point to obtain the characteristic frequency band energy value Ei of the i-th string (i=1,2,3...N, N is the total number of strings in the combiner box), and finally generate the characteristic energy dataset [E1,E2,E3...EN] of the N strings in the current sampling period.

[0023] The formula for calculating the relative difference rate Di of the strings in step S6 is as follows:

[0024] Where: Di is the relative difference rate of the current characteristic energy of the i-th string; Ei is the characteristic frequency band energy of the i-th string; Eavg is the reference energy of the normal string across the entire domain;

[0025] The essential distinction between interference and fault is as follows: Under global common-mode interference, all Ei changes synchronously, Ei and Eavg are basically equal, and Di approaches 0; under single-branch arc fault, the fault string Ei shifts significantly, and Di increases dramatically. In a further technical solution of this invention, the discrimination mechanism of the two-level threshold multi-cycle steady-state verification specifically involves presetting two fixed thresholds: the first level is a suspected early warning threshold, and the second level is a fault confirmation threshold. The specific discrimination logic is as follows:

[0026] (1) Suspected fault warning: If the relative difference rate Di of a certain string is greater than K1, and the abnormal state is maintained for two or more consecutive sampling cycles, it is judged as a suspected DC arc fault, triggering a cloud warning, recording the abnormal string number and the abnormal time period, and not triggering a local alarm.

[0027] (2) Deterministic fault diagnosis: Based on the suspected warning, if the relative difference rate Di of the string continues to be greater than the fault diagnosis threshold K2 and is maintained for 3 or more consecutive sampling cycles, and instantaneous interference and data fluctuation factors are excluded, it is formally determined that the string has a DC series arc fault.

[0028] The present invention provides a photovoltaic array DC arc fault detection system based on string current difference characteristics, characterized in that it specifically includes a combiner box main control unit, a multi-channel DC current sampling module, a synchronous clock module, a signal preprocessing module, and a data processing and fault discrimination module;

[0029] The multi-channel DC current sampling module includes multiple sampling sub-modules, the same number as the number of photovoltaic strings connected to the combiner box. Each sampling sub-module is connected in series in the DC output circuit of a single photovoltaic string and is responsible for collecting the DC current of a single string. All sampling sub-modules are connected to a synchronous clock module to achieve clock synchronization.

[0030] The synchronization clock module provides a unified clock signal for all sampling sub-modules, ensuring that all string current signals are acquired synchronously and aligned in timing.

[0031] The signal preprocessing module receives the raw time-domain current data transmitted by all sampling word modules, integrates a low-pass filtering algorithm to filter out instantaneous spike noise in the line, and outputs a smoothed effective time-domain current signal.

[0032] The main control unit of the combiner box serves as the core control unit, which coordinates and schedules the entire process of sampling, filtering, and calculation, and issues sampling commands and forwards data.

[0033] The data processing and fault diagnosis module incorporates an FFT calculation unit, a characteristic frequency band energy calculation unit, a string difference rate calculation unit, and a multi-level threshold discrimination unit. The FFT calculation unit performs a Fast Fourier Transform on each preprocessed time-domain current sequence to obtain continuous frequency domain spectrum distribution data from 0Hz to 500Hz. The characteristic frequency band energy calculation unit calculates the total fault characteristic frequency band energy of a single string within the characteristic frequency band of 50Hz to 300Hz using a spectrum energy integration algorithm, generating a characteristic energy dataset of multiple strings in the current sampling period, and then filters it. The string difference rate calculation unit calculates the relative difference rate calculation model for each string as the core evaluation index. The multi-level threshold discrimination unit uses a two-level threshold + multi-cycle steady-state verification mechanism to determine the string that ultimately experiences a DC series arc fault.

[0034] The preferred technical solution of the present invention is as follows: The system further includes an alarm output module; the alarm output module includes a local audible and visual alarm and a remote communication interface. After fault determination, it outputs a local alarm signal to trigger an audible and visual fault alarm and remind on-site operation and maintenance personnel; remotely, it uploads the fault type, fault string location, fault occurrence time, and fault characteristic data to the photovoltaic power station operation and maintenance platform via 4G / Ethernet communication to achieve accurate fault location and real-time source tracing, while continuously monitoring the fault status until the fault is eliminated and the data returns to normal.

[0035] The preferred technical solution of the present invention is as follows: the sampling frequency in the synchronous clock module is fixed at 1kHz, and the single-segment sampling duration is strictly controlled to 1 second.

[0036] The core innovative mechanism of this invention is based on three major physical characteristics of photovoltaic strings: homogeneity of operating conditions, common-mode interference, and heterogeneity of faults. It fundamentally solves the problems of poor anti-interference and difficulty in identifying weak faults in traditional frequency domain detection technology. The specific principle is as follows:

[0037] Same operating conditions: All photovoltaic strings connected to the same combiner box have completely identical installation scenarios, light intensity, ambient temperature, wind speed shading, and grounding conditions. Under normal and fault-free operation, the time-domain fluctuation trend, amplitude range, frequency domain spectrum distribution, and harmonic energy ratio of the DC current output by each string are highly consistent, and there are no continuous or significant differences between the strings.

[0038] Common-mode interference characteristics: All conventional interferences at photovoltaic power plant sites are global common-mode interferences, including inverter high-frequency switching harmonic interference, outdoor global electromagnetic radiation interference, instantaneous fluctuations in sunlight interference, and grid coupling noise interference. This type of interference acts synchronously on all photovoltaic strings in the same combiner box, causing the time-domain and frequency-domain characteristics of the current in all strings to change synchronously, in the same direction, and with the same amplitude. The relative characteristic difference between strings remains at an extremely low level, and no differentiated characteristics are produced.

[0039] Anomaly characteristics of faults: Photovoltaic DC series arc faults are local faults in a single branch, occurring only in the series circuits of cables, connectors, terminals, etc. of a single photovoltaic string. They only cause current waveform distortion, high-frequency harmonic surges, and abnormal energy shifts in characteristic frequency bands in the faulty string, while the characteristics of other normal strings remain stable, ultimately forming significant differences in characteristics between strings.

[0040] Based on the above characteristics, this invention abandons the traditional "single-channel signal absolute feature threshold discrimination" logic and instead adopts the "multi-channel string relative difference feature discrimination" logic. It can accurately distinguish between interference and faults that cannot be distinguished by the essential differences between common mode and heterogeneous mode. At the same time, it uses the stable features of normal strings as a reference benchmark to amplify the small distortion features of early weak arcs, so as to achieve accurate fault identification in low signal-to-noise ratio scenarios.

[0041] The multi-channel microsecond-level synchronous sampling technology in this invention achieves synchronous acquisition of the entire string through a unified clock source, ensuring complete consistency in the timing dimension of current data across all branches. This solves the technical pain point of asynchronous sampling's inability to perform horizontal comparisons and is a core prerequisite for achieving difference feature detection; without synchronous sampling, accurate difference calculation is impossible. The adaptive reference energy update mechanism in this invention abandons the traditional fixed threshold discrimination method, dynamically updating the reference energy of the entire string based on real-time operating conditions. This perfectly adapts to the dynamic changes in photovoltaic current with illumination and temperature, avoiding misjudgments of normal operating condition fluctuations as faults and significantly improving the system's environmental adaptability. The common-mode interference shielding mechanism in this invention uses string difference features to replace single-path absolute features, preventing all global common-mode interference from generating string differences. From an algorithmic perspective, this completely shields common industry interference problems such as inverter switching frequency interference, environmental electromagnetic noise, and global illumination fluctuations, reducing the false alarm rate at its source. The weak fault feature amplification mechanism in this invention uses the stable features of a large number of normal strings as a reference, which can accurately capture the small energy shift of early arc faults and transform the weak fault features that are submerged by noise into significant quantitative difference indicators, thus solving the technical problem that traditional single-channel detection cannot identify early minor arc faults.

[0042] This invention features a low overall computational load and strong real-time performance. The entire computation process can be implemented in conventional embedded MCUs and DSP processors, eliminating the need for high-end computing equipment and adapting to the hardware configurations of photovoltaic combiner boxes and local monitoring devices. It employs a cyclic iterative computation mode, completing a full-process detection every second, with real-time data updates and fault identification. Balancing detection accuracy and real-time performance, it can operate stably for extended periods in complex and harsh outdoor conditions involving high temperatures, low temperatures, and high interference, without algorithm failure or data lag issues.

[0043] Based on extensive experimental data on photovoltaic DC series arc faults, it is known that the effective fault characteristics of photovoltaic DC series arcs are concentrated in the low-to-mid-frequency range, while inverter switching interference and environmental stray noise are mostly distributed in a fixed high-frequency range. This invention specifically defines the arc characteristic frequency band as 50Hz~300Hz. This frequency band can completely cover the arcing harmonic characteristics of DC series arcs, while avoiding interference from most fixed-frequency equipment, providing a dedicated frequency domain range for accurate extraction of fault characteristics.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] (1) The detection method in this invention is changed from "independent discrimination of a single string" to "lateral difference comparison and discrimination of multiple strings". Existing technology only performs frequency domain analysis and fixed threshold judgment on the current of a single string, which cannot distinguish between global interference and fault characteristics. This invention relies on the common characteristics of string operating conditions under the same combiner box. Inverter switching harmonics, outdoor light disturbances, global electromagnetic noise and other interference signals are common to all strings. The energy of each string frequency band changes synchronously. The string difference rate is always in the normal range. The device will not trigger an alarm. Only the actual DC series arc fault will produce branch differences. This invention uses the relative difference between strings as the core fault basis, rather than the absolute amplitude of a single signal. Global interference acts on all strings. There is no obvious difference between strings. It is directly filtered out. Only when an arc fault occurs in a single branch will a significant difference be produced. It distinguishes interference from fault from the root and effectively filters out various conventional interferences. The measured false alarm rate is reduced by more than 85% compared with the traditional single-channel frequency domain scheme.

[0046] (2) The present invention adopts global synchronous sampling instead of distributed asynchronous sampling; the traditional scheme has independent sampling of each group of strings and asynchronous clocks, making it impossible to carry out accurate horizontal comparison; the present invention adopts multi-channel synchronous acquisition with a unified clock source, a sampling frequency of more than 1kHz and a fixed time window of 1 second, to ensure that multiple data can be horizontally compared, providing a reliable data source for the calculation of difference features.

[0047] (3) This invention adopts a dual criterion of characteristic frequency band energy + string difference rate; abandons the single frequency domain energy threshold criterion, and combines the characteristic frequency band energy extracted by FFT with the string relative difference rate. For series arc faults with weak initial arcing and current distortion, the single frequency domain signal is easily submerged by noise; while the difference characteristics between the fault string and the surrounding normal string will be amplified. The dual criteria can stably identify early faults, retain the ability of frequency domain analysis to identify arc harmonic characteristics, and strengthen the fault identification through the difference rate. It takes into account both weak fault identification and anti-interference ability, solves the problem of missed detection in traditional schemes, eliminates fire safety hazards in advance, and improves the detection capability of early weak series arc faults.

[0048] (4) This invention enables precise location of faulty strings. During the detection process, the difference index of each string can be calculated in real time. Once a fault is determined, the abnormal string number can be directly locked without the need for a full-site check, which greatly reduces the workload of on-site maintenance and improves the efficiency of fault handling.

[0049] (5) This invention uses the mature FFT frequency domain analysis algorithm and only adds multi-channel synchronous acquisition and difference calculation logic. It can be directly compatible with existing photovoltaic combiner boxes and current sampling hardware. There is no need to replace the field equipment on a large scale. It is easy to promote and apply in batches in existing photovoltaic power plants and new power plants. It has strong compatibility and low transformation cost.

[0050] (6) The present invention has high operational stability, fixed sampling parameters (≥1kHz sampling frequency, 1 second time window), simple algorithm logic, small amount of computation, and can be stably run by embedded processor, adapting to harsh working conditions such as high temperature, low temperature, and high dust outdoors. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the process of the method of the present invention;

[0053] Figure 2 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0055] The present invention will be further described below with reference to embodiments.

[0056] This invention provides a photovoltaic array DC arc fault detection system based on string current difference characteristics, specifically including a combiner box main control unit, a multi-channel DC current sampling module, a synchronous clock module, a signal preprocessing module, a data processing and fault discrimination module, and an alarm output module;

[0057] The multi-channel DC current sampling module includes multiple sampling sub-modules, the same number as the number of photovoltaic strings connected to the combiner box. Each sampling sub-module is connected in series in the DC output circuit of a single photovoltaic string and is responsible for collecting the DC current of a single string. All sampling sub-modules are connected to a synchronous clock module to achieve clock synchronization.

[0058] The synchronous clock module provides a unified clock signal for all sampling sub-modules, ensuring that all string current signals are acquired synchronously and aligned in timing; the sampling frequency is fixed at 1kHz, and the single-segment sampling duration is strictly controlled to 1 second.

[0059] The signal preprocessing module receives the raw time-domain current data transmitted by all sampling word modules, integrates a low-pass filtering algorithm to filter out instantaneous spike noise in the line, and outputs a smoothed effective time-domain current signal.

[0060] The main control unit of the combiner box serves as the core control unit, which coordinates and schedules the entire process of sampling, filtering, and calculation, and issues sampling commands and forwards data.

[0061] The data operation and fault discrimination module is internally provided with an FFT calculation unit, a characteristic frequency band energy calculation unit, a string difference rate calculation unit and a multi-level threshold discrimination unit; the FFT calculation unit performs fast Fourier transform one by one for each preprocessed time-domain current sequence to obtain continuous frequency-domain spectrum distribution data from 0Hz to 500Hz; the characteristic frequency band energy calculation unit calculates the total energy of the fault characteristic frequency band of a single string through a spectrum energy integration algorithm for the frequency-domain data of each string current within the 50Hz~300Hz characteristic frequency band, generates a characteristic energy data set of multiple strings under the current sampling period, and performs filtering processing on the data set; the string difference rate calculation unit calculates the relative difference rate calculation model of each group of strings as the core evaluation index; the multi-level threshold discrimination unit adopts a discrimination mechanism of two-level threshold + multi-cycle steady-state verification to determine the string where the DC series arc fault finally occurs;

[0062] The alarm output module comprises a local acousto-optic alarm and a remote communication interface, outputs a local alarm signal after a fault is determined, and uploads the fault string number and fault type to the operation and maintenance platform of the photovoltaic power station.

[0063] The detection system matched with the present invention adopts a software and hardware cooperative architecture, takes a photovoltaic combiner box as the minimum independent detection unit, is adaptive to all string-type photovoltaic array topologies, and is overall divided into four levels: a hardware acquisition layer, a signal preprocessing layer, a data operation analysis layer and a fault discrimination output layer. All levels cooperate with each other and are synchronized in timing, forming a complete closed-loop detection system.

[0064] Wherein, the hardware acquisition layer: the core is a multi-channel synchronous current sampling unit and a high-precision synchronous clock unit. One-to-one independent sampling channels are configured for all photovoltaic strings under the same combiner box, without channel multiplexing and cross data interference; all sampling channels are bound to a unified clock source to achieve microsecond-level synchronous sampling, ensure that the timing of multiple current signals is completely aligned, provide high-precision raw data support for subsequent horizontal difference comparison, and solve the problems of timing misalignment and data inability to benchmark in traditional asynchronous sampling.

[0065] Signal preprocessing layer: it integrates the functions of adaptive noise reduction, abnormal pulse elimination and data normalization, and accurately corrects the problems of instantaneous spike noise, random pulse interference and amplitude offset in outdoor photovoltaic scenarios. On the premise of completely retaining the high-frequency characteristics of arc faults and the effective distortion characteristics of current, invalid interference signals are filtered out, and the effectiveness and accuracy of raw data are improved.

[0066] Data processing and analysis layer: This is the core algorithm layer of the invention. It integrates a Fast Fourier Transform (FFT) module, an arc-specific characteristic frequency band energy calculation module, a string reference value solving module, a relative difference rate calculation module, and a feature steady-state verification module. It can batch complete the frequency domain conversion, feature quantization, and difference modeling of multiple signals, and realize the accurate conversion from the original time domain signal to fault feature data.

[0067] Fault identification output layer: It adopts a two-level threshold identification logic of "suspected warning + steady-state confirmation" and a multi-window continuous steady-state verification mechanism. Unlike the traditional single threshold instantaneous identification method, it can effectively avoid misjudgment caused by short-term random disturbances and instantaneous data anomalies, accurately output fault warning, fault confirmation and fault location results, and simultaneously realize local alarm and remote data upload.

[0068] The embodiment provides a photovoltaic array DC arc fault detection method based on string current difference characteristics. This method is a cyclic real-time detection system with a fixed single-cycle detection duration of 1 second. The entire process is automated and requires no manual intervention. Detailed steps are as follows:

[0069] S1: Equipment installation and parameter initialization; complete hardware deployment in each combiner box of the photovoltaic array, using each combiner box as an independent detection unit, and set up N sampling sub-modules for all N photovoltaic strings connected to it (N≥2, typically 4, 8, or 12 strings), connecting each of the N sampling sub-modules to the DC circuit of each string; complete parameter configuration after power-on: uniformly set the sampling frequency to 1kHz and the single sampling time window to 1s; based on the photovoltaic DC arc spectrum test data of this region, define the arc-specific characteristic frequency band (a fixed frequency band can be preset according to the power station operating conditions).

[0070] Two threshold levels are preset: a difference warning threshold K1 and an arc fault judgment threshold K2 (K2 > K1). The synchronous clock module completes time calibration to ensure that the clock error of all sampling modules is less than 1μs. Unified hardware and algorithm parameters establish the foundation for synchronous sampling and unified discrimination, completely eliminating phase deviation and data misalignment problems caused by asynchronous sampling. The raw time-domain data of DC current in each string is continuously collected to form N-channel raw current datasets with consistent dimensions and time sequence alignment, which are synchronously transmitted to the preprocessing module.

[0071] S2: Multi-channel synchronous acquisition of time-domain current signals; the synchronous clock module triggers the sampling command, and all current sampling modules start acquisition simultaneously, continuously acquiring DC current time-domain data for 1 second; after the 1-second sampling window ends, all modules immediately stop sampling, and the multiple raw current data streams are transmitted in parallel to the signal preprocessing module. Key points of this step: strict synchronous acquisition to eliminate timing deviations caused by sequential sampling, ensuring that subsequent data sets have comparative value.

[0072] S3: Signal Preprocessing (Filtering and Noise Reduction); To eliminate invalid interference under complex outdoor operating conditions and retain complete arc fault characteristics, layered preprocessing is performed on each original time-domain current signal. First, a first-order Butterworth low-pass filter algorithm is used to filter out irregular noise such as ultra-high frequency instantaneous spikes above 10kHz, electrostatic interference, and instantaneous electromagnetic pulses from equipment, avoiding abnormal pulse interference in subsequent spectrum analysis. Second, baseline calibration is performed to eliminate current amplitude baseline offset problems caused by sampling hardware temperature drift and line voltage drop. Third, single-point abnormal extreme values ​​within a single frame of data are removed, retaining continuous and stable effective current waveforms. The preprocessed signal outputs a standardized time-domain current signal with a significantly improved signal-to-noise ratio, ensuring complete uniformity of processing algorithms and parameters for each signal, guaranteeing horizontal comparability of the data. The filtered signals are then sent to the data processing and fault diagnosis modules respectively. The purpose of this step is to eliminate irregular instantaneous noise and prevent noise interference with subsequent spectrum analysis results.

[0073] S4: Perform FFT transformation on each channel to extract energy in the characteristic frequency band; the data processing and fault diagnosis module performs Fast Fourier Transform (FFT) on each filtered time-domain current signal to convert the one-dimensional time-domain sequence into a frequency-domain amplitude sequence; the discrete time-domain current signal is completely converted into a discrete frequency-domain amplitude signal to obtain continuous frequency-domain spectrum distribution data from 0Hz to 500Hz; combined with a large amount of photovoltaic DC series arc fault test data, it can be seen that the effective fault characteristics of photovoltaic DC series arc are concentrated in the low-frequency to mid-frequency band, while inverter switching interference and environmental stray noise are mostly distributed in a fixed high-frequency band.

[0074] For the frequency domain data of each string current, within the characteristic frequency band of 50Hz~300Hz, the total energy of the fault characteristic frequency band of a single string is calculated using the spectrum energy integration algorithm. Specifically, the calculation method is as follows: traverse the amplitude parameters of all frequency points within the characteristic frequency band, accumulate and integrate the squares of the amplitudes at each frequency point to obtain the characteristic frequency band energy value Ei of the i-th string (i=1,2,3...N, where N is the total number of strings in the combiner box). The specific calculation formula is as follows:

[0075]

[0076] in, is the total energy of the arc characteristic frequency band of the i-th photovoltaic string, in J; i is the string number, with values ​​i=1, 2, 3…N, where N represents the total number of photovoltaic strings connected to the current combiner box; this parameter is used to quantify the degree of frequency domain distortion of the current of a single string, and is the core basic quantitative indicator for identifying arc faults. When an arc fault occurs in a string, this value will undergo a specific abnormal shift.

[0077] The frequency points are the frequency domain sampling points after the FFT transformation, with the unit being Hz, representing the frequency components of the current signal after decomposition.

[0078] The lower limit frequency of the arc characteristic frequency band is fixed at 50Hz in this invention, which is the lowest starting frequency of the effective fault characteristics of DC series arc.

[0079] The upper limit frequency of the arc characteristic frequency band is 300Hz, which is the highest termination frequency of the effective fault characteristics of DC series arc.

[0080] For the i-th string time-domain current signal after FFT transformation, the complex amplitude at the corresponding frequency f is ; The modulus of the complex number represents the magnitude of the current harmonic amplitude at the corresponding frequency point;

[0081] This formula is a standard and commonly used formula in the fields of electrical signal processing and fault spectrum analysis. Its principle is to accumulate the squares of the harmonic amplitudes of all effective frequency points within the characteristic frequency band, accurately aggregating the high-frequency harmonic energy corresponding to the arc fault. This transforms the weak current waveform distortion, invisible to the naked eye, into quantifiable and comparable standardized energy values. Through this step, the difficult-to-quantify arc waveform distortion characteristics are transformed into quantifiable and comparable energy values, achieving the datafication and standardization of fault characteristics, ultimately generating a characteristic energy dataset [E1, E2, E3...EN] for N-channel strings in the current sampling period. This step specifically defines the arc characteristic frequency band as 50Hz~300Hz. This frequency band can completely cover the arc harmonic characteristics of DC series arcs, while avoiding interference from most fixed-frequency equipment, providing a dedicated frequency domain range for accurate fault feature extraction. The key point of this step is to use the mature FFT algorithm to extract arc frequency domain features, ensuring that the basic arc features are not lost; this is the core step in fault feature extraction.

[0082] S5: Calculate the baseline energy and relative difference rate of the string; remove the extreme values ​​(single instantaneous disturbance data) that are obviously abnormal in the data set, and calculate the average frequency band energy of the remaining normal strings, which is denoted as the baseline energy Eavg;

[0083] For each string, calculate the relative difference rate Di according to the formula:

[0084]

[0085] The difference rates of all strings are obtained sequentially [D1, D2, D3...Dn].

[0086] This step is the most crucial part of the entire solution: If it is common-mode interference such as inverter switching noise or global environmental noise: all Ei increases or decreases synchronously, Ei and Eavg are basically equal, Di≈0, and there is no significant difference in string groups;

[0087] If a DC series arc fault occurs in a certain string: the Ei of the faulty string is distorted, the deviation from Eavg increases significantly, and Di increases significantly, forming obvious abnormal characteristics. Through this step, the essential distinction between interference and fault can be achieved.

[0088] S6: Multi-level threshold fault detection: Level 1 detection (early warning judgment): Traverse all string difference rates Di. If a string Di > preset early warning threshold K1, and this state lasts for 2 consecutive sampling windows, it is determined to be a suspected arc fault, triggers an early warning signal, and uploads the suspected fault string number.

[0089] Secondary discrimination (fault determination): If the Di of the string continues to be greater than the fault determination threshold K2, and the abnormal state is maintained for 3 or more consecutive sampling windows, excluding the influence of instantaneous disturbances, the string is formally determined to have a photovoltaic DC series arc fault.

[0090] The purpose of setting up multi-level, multi-window continuous judgment is to further filter short-term random disturbances, avoid misjudgment caused by single data anomalies, and improve detection reliability.

[0091] S7: Fault result output and operation and maintenance linkage;

[0092] Suspected warning status: The alarm output module only uploads warning information to the operation and maintenance platform, and does not trigger audible and visual alarms locally;

[0093] Confirm arc fault status: Locally activate audible and visual alarms, and simultaneously upload the fault type, fault string number, and fault occurrence time to the power plant remote operation and maintenance platform via the communication interface to remind operation and maintenance personnel to conduct on-site repairs in a timely manner.

[0094] This invention provides a typical operating condition description for specific applications.

[0095] Under normal operating conditions: the illumination and operating conditions of all strings in the same combiner box are consistent, the Ei of each string are basically close, all Di are much smaller than K1, the system has no alarms, and the "sampling-analysis-discrimination" process is continuously executed in a loop;

[0096] Full-domain interference condition (inverter switch, ambient electromagnetic noise): All strings Ei are synchronously offset, but the relative difference between strings remains unchanged, Di is always normal, and the system does not alarm;

[0097] Single-string DC series arc fault condition: When the fault string Ei is distorted and Di exceeds K1 and K2, the system outputs early warning and fault alarm in sequence, and locks the fault string to complete fault detection and location.

[0098] The above description is merely one embodiment of the present invention, and while it is detailed and specific, it should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for detecting DC arc faults in photovoltaic arrays based on string current difference characteristics, characterized in that, Specifically, the following steps are included: S1. Multi-channel synchronous current signal acquisition: Using a single combiner box as an independent detection unit, full-channel synchronous data acquisition is performed on all N photovoltaic strings connected to it. The raw time-domain data of DC current of each string is continuously acquired to form N raw current datasets with consistent dimensions and time sequence alignment, which are synchronously transmitted to the preprocessing module, where N≥2. S2. Layered signal preprocessing and data correction: After layered signal preprocessing of each raw current data collected in step S1, the preprocessing module outputs a standardized time-domain current signal with a significantly improved signal-to-noise ratio, and ensures that the processing algorithms and parameters of each signal are completely consistent, so as to ensure that the data has horizontal comparability. S3. Full-channel signal FFT frequency domain conversion and characteristic frequency band delineation: For each pre-processed time-domain current sequence, perform fast Fourier transform one by one to completely convert the discrete time-domain current signal into a discrete frequency-domain amplitude signal, and obtain continuous frequency domain spectrum distribution data from 0Hz to 500Hz. S4. Precise calculation of energy in the characteristic frequency band of a single string; For the frequency domain data of the current of each string, the total energy of the fault characteristic frequency band of a single string is calculated in the characteristic frequency band of 50Hz~300Hz through the spectrum energy integration algorithm, and the characteristic energy dataset of N strings under the current sampling period is generated. S5. Solving string baseline energy and filtering outlier data; The feature energy dataset generated in step S4 is preprocessed and filtered to remove instantaneous extreme values ​​that deviate from the mean by more than 50% within a single cycle; then the arithmetic mean of the energy values ​​of the remaining valid normal strings is calculated and defined as the global baseline energy Eavg under the current operating conditions of the combiner box. S6. Modeling and Calculation of Relative Difference Rate of Strings; Constructing a relative difference rate calculation model of strings as the core evaluation index, and quantifying the degree of feature deviation of a single string by calculating the relative difference rate of each string to achieve the essential distinction between interference and fault; S7. Multi-level steady-state threshold fault discrimination: A discrimination mechanism of two-level threshold + multi-cycle steady-state verification is adopted. When the relative difference rate of the string group is continuously greater than the fault diagnosis threshold and is maintained for 3 or more consecutive sampling cycles, it is formally determined that the string group has a DC series arc fault.

2. The method for detecting DC arc faults in photovoltaic arrays based on string current difference characteristics according to claim 1, characterized in that, The method further includes: S8. Fault Location and Hierarchical Output: After fault diagnosis, the system automatically locks the unique number of the abnormal string and synchronously executes hierarchical output: Locally triggers audible and visual fault alarms to remind on-site maintenance personnel; Remotely, through 4G / Ethernet communication, it uploads the fault type, fault string location, fault occurrence time, and fault characteristic data to the photovoltaic power station operation and maintenance platform to achieve accurate fault location and real-time source tracing, while continuously monitoring the fault status until the fault is eliminated and the data returns to normal.

3. A method for detecting DC arc faults in a photovoltaic array based on string current difference characteristics, as described in claim 1 or 2, characterized in that... In step S1, all sampling channels have consistent sampling accuracy and response speed, and the sampling parameters are uniformly configured: the sampling frequency is fixed at 1kHz or higher, the single-frame sampling time window is strictly locked at 1s, and the number of sampling points per channel per time is not less than 1000 to ensure the integrity of the time domain signal; all sampling channels share the same GPS or local high-precision synchronization clock, and the sampling timing error between channels is ≤1μs, completely eliminating the phase deviation and data misalignment problems caused by asynchronous sampling.

4. A method for detecting DC arc faults in a photovoltaic array based on string current difference characteristics, as described in claim 1 or 2, characterized in that, The S2 step-by-step layered preprocessing specifically includes: First, using a first-order Butterworth low-pass filter algorithm to filter out irregular noise above 10kHz, specifically including ultra-high frequency instantaneous spike pulses, electrostatic interference, and instantaneous electromagnetic pulses from equipment; Second, performing baseline calibration processing to eliminate the current amplitude baseline offset problem caused by sampling hardware temperature drift and line voltage drop; Third, removing single-point abnormal extreme values ​​within a single frame of data and retaining continuous and stable valid current waveforms.

5. A method for detecting DC arc faults in a photovoltaic array based on string current difference characteristics, as described in claim 1 or 2, characterized in that: In step S3, a 1-second time-domain current sequence is preprocessed for each channel.

6. A method for detecting DC arc faults in a photovoltaic array based on string current difference characteristics, as described in claim 1 or 2, characterized in that, The specific calculation method in step S4 is as follows: traverse the amplitude parameters of all frequency points in the characteristic frequency band, accumulate and integrate the squares of the amplitude of each frequency point to obtain the characteristic frequency band energy value Ei of the i-th string, i=1,2,3...N, where N is the total number of strings in the combiner box, and finally generate the characteristic energy dataset of N strings in the current sampling period [E1,E2,E3...EN]; The formula for calculating the relative difference rate Di of the strings in step S6 is as follows: ； Where: Di is the relative difference rate of the current characteristic energy of the i-th string; Ei is the characteristic frequency band energy of the i-th string; Eavg is the reference energy of the normal string across the entire domain; The essential distinction between interference and fault is as follows: Under global common-mode interference, all Ei changes synchronously, Ei and Eavg are basically equal, and Di approaches 0; under single-branch arc fault, the fault string Ei shifts significantly, and Di increases dramatically.

7. A method for detecting DC arc faults in a photovoltaic array based on string current difference characteristics, as described in claim 1 or 2, characterized in that, The discrimination mechanism of two-level thresholds + multi-cycle steady-state verification in step S7 specifically involves pre-setting two fixed thresholds: the first level is the suspected early warning threshold K1, and the second level is the fault confirmation threshold K2, where K2 > K1. The specific discrimination logic is as follows: (1) Suspected fault warning: If the relative difference rate Di of a certain string is greater than K1, and the abnormal state is maintained for two or more consecutive sampling cycles, it is judged as a suspected DC arc fault, triggering a cloud warning, recording the abnormal string number and the abnormal time period, and not triggering a local alarm. (2) Deterministic fault diagnosis: Based on the suspected warning, if the relative difference rate Di of the string continues to be greater than the fault diagnosis threshold K2 and is maintained for 3 or more consecutive sampling cycles, and instantaneous interference and data fluctuation factors are excluded, it is formally determined that the string has a DC series arc fault.

8. A photovoltaic array DC arc fault detection system based on string current difference characteristics, characterized in that: Specifically, it includes a combiner box main control unit, a multi-channel DC current sampling module, a synchronous clock module, a signal preprocessing module, and a data processing and fault diagnosis module; The multi-channel DC current sampling module includes multiple sampling sub-modules, the same number as the number of photovoltaic strings connected to the combiner box. Each sampling sub-module is connected in series in the DC output circuit of a single photovoltaic string and is responsible for collecting the DC current of a single string. All sampling sub-modules are connected to a synchronous clock module to achieve clock synchronization. The synchronization clock module provides a unified clock signal for all sampling sub-modules, ensuring that all string current signals are acquired synchronously and aligned in timing. The signal preprocessing module receives the raw time-domain current data transmitted by all sampling word modules, integrates a low-pass filtering algorithm to filter out instantaneous spike noise in the line, and outputs a smoothed effective time-domain current signal. The main control unit of the combiner box serves as the core control unit, which coordinates and schedules the entire process of sampling, filtering, and calculation, and issues sampling commands and forwards data. The data processing and fault diagnosis module incorporates an FFT calculation unit, a characteristic frequency band energy calculation unit, a string difference rate calculation unit, and a multi-level threshold discrimination unit. The FFT calculation unit performs a Fast Fourier Transform on each preprocessed time-domain current sequence to obtain continuous frequency domain spectrum distribution data from 0Hz to 500Hz. The characteristic frequency band energy calculation unit calculates the total fault characteristic frequency band energy of a single string within the characteristic frequency band of 50Hz to 300Hz using a spectrum energy integration algorithm, generating a characteristic energy dataset of multiple strings in the current sampling period, and then filters it. The string difference rate calculation unit calculates the relative difference rate calculation model for each string as the core evaluation index. The multi-level threshold discrimination unit uses a two-level threshold + multi-cycle steady-state verification mechanism to determine the string that ultimately experiences a DC series arc fault.

9. A photovoltaic array DC arc fault detection system based on string current difference characteristics according to claim 8, characterized in that: The system also includes an alarm output module; the alarm output module includes a local audible and visual alarm and a remote communication interface. After fault determination, it outputs a local alarm signal to trigger an audible and visual fault alarm and remind on-site maintenance personnel. It also remotely uploads fault type, fault string location, fault occurrence time, and fault characteristic data to the photovoltaic power station operation and maintenance platform via 4G / Ethernet communication to achieve accurate fault location and real-time source tracing, while continuously monitoring the fault status until the fault is eliminated and the data returns to normal.

10. A photovoltaic array DC arc fault detection system based on string current difference characteristics according to claim 8, characterized in that: The sampling frequency in the synchronous clock module is fixed at 1kHz, and the single-segment sampling duration is strictly controlled to 1 second.

Images