A method and system for rapid fault positioning of a photovoltaic power station

Abstract

The embodiment of the present application provides a kind of photovoltaic power station fault fast positioning method and system, belong to electrical measurement and fault detection technical field.The method comprises: obtaining the electrical parameter data of each group string of photovoltaic power station and executes normalization processing in combination with environmental working condition, generates baseline data, and filters suspicious group string based on baseline data;The preset power perturbation signal is applied to the suspicious group string, and the corresponding voltage and current response are collected to form dynamic response data;Determine impedance spectrum characteristics based on the dynamic response data and extract dynamic response fingerprint, construct feature vector for representing the electrical state of group string;The feature vector is compared with the health template and phase coupling analysis is performed, to generate fault determination result and output corresponding group string positioning information.The present application scheme realizes the accurate identification and fast positioning of group string level fault of photovoltaic power station, and can still maintain high robustness and high confidence diagnosis effect under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of electrical measurement and fault detection technology, specifically to a method for rapid fault location in a photovoltaic power station and a system for rapid fault location in a photovoltaic power station. Background Technology

[0002] During the operation of large-scale photovoltaic power plants, the number of photovoltaic modules and their strings is enormous and widely distributed. Any single point of failure can lead to local power loss or even affect the output stability of the entire string. To maintain the efficient operation of the power plant, it is necessary to periodically detect and locate anomalies in the operating status of each string. Currently, the widely used monitoring methods mainly include voltage and current sampling monitoring based on combiner boxes or inverter ports, and on-site inspection methods based on infrared thermal imaging. The former relies on changes in steady-state electrical parameters to identify faults, but in the early stages of minor faults such as early module mismatch, bypass diode degradation, or poor contact, the changes in electrical parameters are extremely small and are often masked by the overall averaging effect, making it difficult to detect accurately in the early stages. The latter can identify abnormal modules through temperature distribution, but it relies on manual or drone data collection, is greatly affected by weather, lighting, and viewing angle, cannot achieve continuous online detection, and its positioning accuracy is limited by image resolution and spatial registration error.

[0003] Furthermore, existing electrical parameter monitoring schemes are generally based on passive observation, which involves collecting steady-state voltage, current, and other data under normal operating conditions for comparative analysis. The detection sensitivity of this method is significantly affected by fluctuations in operating conditions. When irradiance or temperature changes rapidly, systematic disturbances are easily misinterpreted as fault signals; while under stable operating conditions, if a fault only causes transient responses or changes in dynamic characteristics, traditional monitoring methods cannot detect it. Due to the lack of quantitative measurement methods for the dynamic behavior of electrical parameters, existing schemes often only locate the fault at the busbar level, failing to achieve accurate identification at the string or even module level.

[0004] Therefore, the core issue in the field of photovoltaic power plant fault detection is how to accurately identify and locate early or minor electrical anomalies through more sensitive observation of string electrical responses without shutting down the system. Existing technologies still have significant shortcomings in signal driving methods, dynamic feature extraction, and positioning accuracy, leading to problems such as susceptibility to external operating conditions, insignificant response characteristics, and coarse fault location granularity. There is an urgent need to propose a string-level fault detection and rapid location method that can be executed online, has high sensitivity, and provides accurate positioning. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for rapid fault location in photovoltaic power plants, so as to at least solve the problems of low fault detection sensitivity, susceptibility to operating condition fluctuations, and difficulty in achieving string-level accurate location in the prior art.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for rapid fault location in a photovoltaic power station. The method includes: acquiring electrical parameter data of each string in the photovoltaic power station and performing normalization processing in conjunction with environmental conditions to generate baseline data, and screening suspicious strings based on the baseline data; applying a preset power perturbation signal to the suspicious strings and collecting the corresponding voltage and current responses to form dynamic response data; determining impedance spectrum characteristics based on the dynamic response data and extracting dynamic response fingerprints to construct feature vectors for characterizing the electrical state of the strings; comparing the feature vectors with a health template and performing phase coupling analysis to generate fault determination results and output corresponding string location information.

[0007] Optionally, the electrical parameter data of each string in the photovoltaic power station is acquired and normalized in combination with environmental conditions to generate baseline data. This includes: collecting voltage and current signals of each string within a preset sampling period, as well as corresponding irradiance, ambient temperature, and wind speed parameters, and constructing a sampling set containing multi-dimensional operating condition parameters; calculating the operating condition correction coefficient of each string based on the sampling set, and performing normalization operations on the voltage and current data in the sampling set to form baseline data reflecting the output characteristics of each string under the same operating conditions.

[0008] Optionally, the rule for screening suspicious strings based on baseline data is as follows: calculate three normalized indices—voltage deviation, current deviation, and power stability—in the baseline data of each string, and perform sliding window statistical analysis on the normalized indices within a continuous sampling period; when the average absolute value of the voltage deviation or current deviation of any string exceeds a preset threshold, and the power stability is lower than the lower limit of the mean of neighboring strings, or when a trend shift occurs in adjacent time windows, the string is determined to be a suspicious string.

[0009] Optionally, a preset power perturbation signal is applied to the suspected string from the inverter side within a preset amplitude and preset frequency range, and the corresponding instantaneous voltage and current values ​​are synchronously collected during the duration of the perturbation. After detrending processing and operating condition compensation, dynamic response data is generated, including: applying a periodic perturbation to the output power reference value of the target string from the maximum power point tracking channel on the inverter side within a preset amplitude and preset frequency range, and recording the corresponding perturbation control signal; synchronously sampling the instantaneous voltage and current values ​​of the target string during the duration of the perturbation to form a voltage response sequence and a current response sequence; performing detrending processing and operating condition compensation on the voltage response sequence and current response sequence to eliminate the common-mode influence of irradiance and temperature, and aligning them with the perturbation control signal on the time axis to generate corrected dynamic response data.

[0010] Optionally, detrending and operating condition compensation are performed on the voltage response sequence and current response sequence to eliminate the common-mode effects of irradiance and temperature. This includes: extracting the corresponding irradiance change curve and ambient temperature change curve during the duration of the perturbation, and calculating the instantaneous slope coefficient of irradiance on current and the correction coefficient of temperature on voltage; constructing an operating condition compensation model based on the slope coefficient and correction coefficient, and performing linear detrending and differential normalization on the voltage response sequence and current response sequence, respectively.

[0011] Optionally, based on the dynamic response data, impedance spectrum characteristics are determined and dynamic response fingerprints are extracted to construct a feature vector characterizing the string electrical state. This includes: performing a fast Fourier transform on the voltage response sequence and current response sequence after operating condition compensation to calculate the complex impedance spectrum, and extracting corresponding impedance spectrum feature parameters based on the complex impedance spectrum; wherein the impedance spectrum feature parameters include any one or more of impedance magnitude, impedance phase angle, equivalent series resistance, and parasitic capacitance; extracting dynamic fingerprint features in the frequency domain and time domain of the dynamic response data respectively; wherein the dynamic fingerprint features include any one or more of main harmonic amplitude, phase delay, harmonic energy ratio, response time constant, and nonlinear exponent; concatenating the impedance spectrum feature parameters and dynamic fingerprint features in a preset order to form a feature vector, which serves as the feature vector characterizing the string electrical state.

[0012] Optionally, the impedance spectrum feature parameters and dynamic fingerprint features are concatenated into a feature vector in a preset order to represent the string electrical state. This includes: sorting the impedance spectrum feature parameters and dynamic fingerprint features according to their sensitivity to the string electrical state, generating a feature index table to indicate the concatenation position; performing normalization and dimension unification processing on the impedance spectrum feature parameters and dynamic fingerprint features respectively, and writing them sequentially into the same data structure according to the order of the feature index table to form a feature vector representing the string electrical state.

[0013] Optionally, the feature vector is compared with the health template and phase coupling analysis is performed to generate a fault determination result and output the corresponding string location information. This includes: reading the health template vector corresponding to the target string from the historical operation file; performing distance measurement calculation on the current feature vector and the health template to obtain the deviation index of each feature component; calculating the response phase correlation coefficient between the feature vectors of each string within the same busbar unit range to construct a phase coupling matrix; when the deviation of any string exceeds a preset threshold and its correlation in the phase coupling matrix deviates significantly from the health template, the string is determined to be an abnormal object; mapping the abnormal object to the actual physical location according to the power station wiring log, and outputting a location result containing the string number, fault type, and confidence level.

[0014] Optionally, within the same bus unit, the response phase correlation coefficient between the feature vectors of each string is calculated to construct a phase coupling matrix, including: extracting the corresponding phase delay feature components from the feature vectors of each string, and calculating the phase change sequence of adjacent sampling periods within a preset time window; performing correlation coefficient calculation on the phase change sequences of any two strings to obtain phase correlation parameters, and arranging all phase correlation parameters according to the string number to generate a phase coupling matrix.

[0015] A second aspect of this invention provides a rapid fault location system for a photovoltaic power station. The system is used to execute the aforementioned rapid fault location method for a photovoltaic power station. The system includes: a data acquisition unit, used to acquire electrical parameter data of each string in the photovoltaic power station and perform normalization processing in conjunction with environmental conditions to generate baseline data, and to filter suspicious strings based on the baseline data; a feedback unit, used to apply a preset power perturbation signal to the suspicious strings from the inverter side within a preset amplitude and preset frequency range, and to synchronously acquire the corresponding instantaneous voltage and current values ​​during the duration of the perturbation, and to form dynamic response data after detrending processing and operating condition compensation; a processing unit, used to determine impedance spectrum characteristics based on the dynamic response data and extract dynamic response fingerprints to construct feature vectors characterizing the electrical state of the strings; and an output unit, used to compare the feature vectors with a health template and perform phase coupling analysis to generate fault determination results and output corresponding string location information.

[0016] Through the above technical solution, this invention achieves high-precision fault identification and location at the photovoltaic string level by combining steady-state and dynamic electrical parameter analysis under continuous operation. By normalizing the operating conditions of the electrical parameter data of each string, the influence of external factors such as irradiance and temperature is eliminated, ensuring data comparability under different operating conditions. After screening suspicious strings based on baseline data, a micro-perturbation signal is applied to actively stimulate their electrical dynamic response, thereby capturing subtle anomalies that traditional steady-state detection cannot reflect. Furthermore, the impedance spectrum characteristics are calculated using the dynamic response, and dynamic fingerprints are extracted, enabling the detection results to sensitively distinguish early faults such as poor contact, bypass diode degradation, and microcracks. Finally, through comparison with a healthy template and phase coupling analysis, automatic identification and precise location mapping of abnormal strings are achieved. The overall solution significantly improves the sensitivity and location accuracy of photovoltaic power plant fault detection, reduces the impact of environmental interference, and can quickly complete anomaly identification and location without affecting grid-connected operation.

[0017] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0019] Figure 1 This is a flowchart of the steps of a method for rapid fault location in a photovoltaic power station provided by one embodiment of the present invention;

[0020] Figure 2 This is a system structure diagram of a photovoltaic power station fault rapid location system provided in one embodiment of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] like Figure 1 As shown, this invention provides a method for rapid fault location in a photovoltaic power station, the method comprising:

[0023] S10 not: Obtain electrical parameter data of each string in the photovoltaic power station and perform normalization processing in combination with environmental conditions to generate baseline data, and screen suspicious strings based on the baseline data.

[0024] Specifically, voltage and current signals of each string are collected within a preset sampling period, along with corresponding irradiance, ambient temperature, and wind speed parameters, to construct a sampling set containing multi-dimensional operating condition parameters. The operating condition correction coefficient of each string is calculated based on the sampling set, and normalization is performed on the voltage and current data in the sampling set to form baseline data reflecting the output characteristics of each string under the same operating conditions.

[0025] Furthermore, the rules for screening suspicious strings based on baseline data are as follows: three normalized indices, namely voltage deviation, current deviation, and power stability, are calculated in the baseline data of each string, and sliding window statistical analysis is performed on the normalized indices within a continuous sampling period; when the average absolute value of the voltage deviation or current deviation of any string exceeds a preset threshold, and the power stability is lower than the lower limit of the mean of the neighboring strings, or when a trend shift occurs in the adjacent time window, the string is determined to be a suspicious string.

[0026] In this embodiment of the invention, during the daily operation of a photovoltaic power station, the number of strings is enormous, and the output power of each string is constantly affected by meteorological conditions, such as changes in irradiance, temperature rise, and wind speed fluctuations. Even if all equipment is working normally, these external factors will still cause natural fluctuations in output voltage and current. Therefore, to accurately identify strings that are truly abnormal in such a complex context, the first step is to ensure that all data are standardized. In other words, the electrical parameters collected under different operating conditions need to be equivalent to a unified set of conditions.

[0027] In practice, a fixed sampling period is first set for each string, such as one minute or five minutes. The shorter the period, the higher the time resolution of the subsequent analysis. The voltage signal V is recorded at each sampling time. raw (t), Current signal I raw Simultaneously, the irradiance G(t), ambient temperature T(t), and wind speed W(t) at that moment are also collected. These quantities together constitute a multi-dimensional working condition sample set. The collected data are only raw observations and do not yet have the meaning of direct horizontal comparison. They need to be transformed into comparable standard state quantities using a modified model.

[0028] Voltage is typically significantly affected by temperature, therefore a correction factor η is introduced for temperature. T Correct it according to the empirical formula:

[0029] ;

[0030] Among them, T ref This is a set reference temperature, such as 25°C. The current is more controlled by the irradiance and can be corrected proportionally.

[0031] ;

[0032] G ref Standard irradiance, for example, 1000 W / m 2 For certain high-wind-speed areas, a coefficient reflecting the impact of wind speed on component temperature rise can be introduced to make the temperature correction more closely match the site conditions. Through this process, voltage and current data collected under different meteorological conditions are brought back to the same operating condition framework.

[0033] Next, we need to use this normalized data to generate a baseline. This is not a simple averaging, but rather an analysis of the overall output stability of the string over time using a sliding window approach. The average output power is calculated for each time window:

[0034] ;

[0035] Then calculate the standard deviation σ of the power within the window.P This leads to the power stability index:

[0036] ;

[0037] In this way, each string has a baseline sequence that varies over time, including both the average output level and the magnitude of output fluctuations.

[0038] With this baseline data, the next step is to use it to screen for suspicious strings. To do this, we first calculate the voltage deviation ΔV = V for each string relative to the reference strings in the same array. corr -V ref Current deviation ΔI=I corr -I ref And for power stability S p Simultaneous monitoring is performed. Then, the changing trends of these indicators are observed over several consecutive sampling windows. Generally, the ΔV and ΔI of a healthy string will fluctuate within a stable, small range, while S... p Approaching 1. If the absolute value of the average voltage deviation or current deviation of a certain string is found to continuously exceed the preset threshold, it indicates that its output behavior has significantly deviated from the group; if at the same time its power stability is much lower than the average value of neighboring strings, or if a unidirectional drift trend appears in a continuous time window, it can be basically determined that there is an anomaly. Whether it is poor contact, component microcracks or local shading, it is possible to detect it in advance at this stage.

[0039] To prevent false alarms caused by sudden fluctuations in operating conditions, these conditions can be met continuously over multiple time windows before a flag is officially triggered. Threshold settings are typically based on historical healthy operating data, such as using the 95th percentile of the deviation distribution during past stable periods as the judgment boundary. This ensures sensitivity without causing frequent false alarms.

[0040] Overall, the processing logic in this stage is to first standardize the data under different environments, and then use statistical methods to identify outlier strings. This significantly reduces common-mode interference caused by meteorological factors, allowing output fluctuations to truly reflect the state of the strings themselves. On the other hand, the range of suspicious strings screened out usually only accounts for a very small portion of the total, saving computational and detection costs for subsequent dynamic stimulus and response analysis.

[0041] Step S20: Apply a preset power perturbation signal to the suspected string from the inverter side within a preset amplitude and preset frequency range, and synchronously collect the corresponding instantaneous voltage and current values ​​during the duration of the perturbation. After detrending processing and operating condition compensation, dynamic response data is formed.

[0042] Specifically, the maximum power point tracking channel on the inverter side applies periodic perturbations to the target string output power reference value within a preset amplitude and frequency range, and records the corresponding perturbation control signal; during the duration of the perturbation, the instantaneous voltage and current values ​​of the target string are simultaneously sampled to form a voltage response sequence and a current response sequence; the voltage response sequence and current response sequence are subjected to detrending processing and operating condition compensation to eliminate the common-mode effects of irradiance and temperature, and are aligned with the perturbation control signal on the time axis to generate corrected dynamic response data.

[0043] Furthermore, detrending and operating condition compensation are performed on the voltage response sequence and current response sequence to eliminate the common-mode effects of irradiance and temperature. This includes: extracting the corresponding irradiance change curve and ambient temperature change curve during the duration of the perturbation, and calculating the instantaneous slope coefficient of irradiance on current and the correction coefficient of temperature on voltage; constructing an operating condition compensation model based on the slope coefficient and correction coefficient, and performing linear detrending and differential normalization on the voltage response sequence and current response sequence, respectively.

[0044] In this embodiment of the invention, after completing the initial baseline screening, dynamic detection needs to be further performed on strings identified as suspicious. This is because many early faults, such as increased contact resistance, degradation of bypass diode performance, and microcracks in components, cannot be fully detected based solely on steady-state data. These anomalies often only become apparent under dynamic disturbances. Therefore, it is necessary to apply a controllable, small-amplitude power disturbance to the suspicious strings, causing their operating point to deviate regularly from the maximum power point over a short period, while simultaneously recording the instantaneous voltage and current responses, thereby generating dynamic response data.

[0045] Specifically, on the inverter side, the maximum power point tracking (MPPT) channel's regulation capability is used to apply a periodic perturbation signal to the target string's output power reference value. The perturbation can be a sinusoidal signal, a square wave signal, or a pseudo-random binary sequence (PRBS), typically a sinusoidal or frequency-sweep form, to facilitate subsequent frequency domain analysis of the response characteristics. The perturbation amplitude is generally set to 0.5% to 2% of the rated power to ensure no significant power generation loss while being sufficient to trigger a dynamic response of the electrical parameters. The frequency range can be selected between 0.1Hz and 3Hz; the lower frequency range is beneficial for stimulating slow-changing processes, while the higher frequency range is suitable for detecting local parasitic effects. The perturbation signal can be represented as:

[0046] ;

[0047] in, For steady-state output power, This is the disturbance amplitude scaling factor. The frequency is the disturbance frequency.

[0048] During the perturbation signal injection, the voltage at the serial port needs to be sampled synchronously. and current The sampling frequency should be at least 10 times the disturbance frequency to ensure accurate capture of the waveform characteristics of the dynamic response. The resulting raw data contains the direct electrical response of the target string to power disturbances, but it also contains low-frequency interference from environmental fluctuations, thus requiring further processing.

[0049] First, detrending is performed on the voltage and current response sequences. The purpose of detrending is to eliminate the DC offset caused by slow changes in illumination or temperature, ensuring the sequence retains only the components that change synchronously with the disturbance. This can be achieved through least-squares linear fitting or moving average filtering. The detrended signals are denoted as follows: and .

[0050] Secondly, operational condition compensation is necessary because even with a short duration of disturbance, irradiance and temperature may still experience slight changes, which can generate common-mode noise in the dynamic signal. Therefore, the irradiance variation curve is recorded synchronously during the perturbation process. and ambient temperature change curve The instantaneous effects of irradiance and current on voltage are calculated. The linear slope coefficient of irradiance with respect to current can be expressed as:

[0051] ;

[0052] The temperature correction factor for voltage can be expressed as:

[0053] ;

[0054] Based on these two coefficients, a working condition compensation model is constructed, and the response sequence is differentially normalized to obtain the corrected voltage response sequence. and current response sequence Its expression is as follows:

[0055] ;

[0056] in, and These represent the initial temperature and irradiance at the start of the disturbance, respectively.

[0057] After the above corrections, low-frequency common-mode interference in the voltage and current sequences is significantly suppressed, leaving mainly the dynamic response components corresponding to the power disturbance. To ensure the timing accuracy of subsequent analysis, the compensated response data needs to be aligned with the disturbance control signal on the time axis. The alignment method can be based on cross-correlation or phase-locked loop algorithms, ensuring that the response waveform of each cycle strictly corresponds to the disturbance phase, thus providing a consistent phase reference for impedance spectrum calculation.

[0058] The significance of this processing method lies in its ability to separate dynamic characteristics that were originally obscured by light fluctuations and temperature drift, making the electrical inertia, nonlinearity, and response delay of each suspect string clearly visible. Through this dynamic response data, not only can the equivalent impedance changes of the string be analyzed, but its transient behavior can also be further revealed, such as whether there is capacitance decay or conduction hysteresis.

[0059] During implementation, the perturbation parameters can be flexibly adjusted according to the site conditions. If the lighting conditions are stable, the perturbation frequency can be appropriately increased to shorten the test time; if the environmental fluctuations are strong, the perturbation amplitude can be reduced and the sampling time extended to improve the signal-to-noise ratio. For the detection of multiple suspicious strings, a time-division injection strategy can be adopted, that is, applying micro-perturbations sequentially to avoid the coupling effect of multiple strings being perturbed simultaneously on the DC bus.

[0060] After completing this entire set of steps, the obtained dynamic response data, corrected for operating conditions, can be directly used for subsequent frequency domain analysis or impedance feature extraction. This method has significant advantages over traditional steady-state detection. Steady-state detection can only observe whether power or current decreases, while the dynamic excitation here can capture subtle deformations in the response curve, even if the fault has not yet caused significant power loss, it can be reflected in phase lag or abnormal response amplitude.

[0061] This process enables repeatable and quantifiable measurements of the dynamic characteristics of photovoltaic strings without shutting down the system, providing a stable input basis for subsequent impedance spectroscopy analysis. By combining perturbation excitation with operating condition compensation, the signal-to-noise ratio of the detected signal is significantly improved, ensuring that effective dynamic response characteristics can be stably extracted even under strong background noise and complex weather conditions.

[0062] In another possible implementation, to further enhance the anti-interference capability of dynamic detection under complex weather conditions, instead of using traditional sinusoidal or frequency-sweeping perturbations, a random pulse envelope perturbation signal is introduced. The principle is to use low-duty-cycle, non-periodic power pulses to cause the string to experience rapid power fluctuations in a very short time, and then analyze the recovery process of the response to determine the electrical characteristics. While this perturbation method appears random, the duration, energy density, and interval of each pulse envelope are actually controlled by a preset strategy, thereby stimulating electrical behaviors with different time constants within the component.

[0063] In practice, each pulse envelope is limited to the millisecond to second range, and multi-level amplitude distributions induce different levels of response in voltage and current. Since the disturbance duration is extremely short, its impact on the overall array's power generation is negligible; however, this brief excitation is sufficient to activate the dynamic changes in parasitic capacitance, junction capacitance, and local contact resistance within the component. Subsequently, during the natural recovery phase after the pulse ends, the decay curves of voltage and current are continuously acquired. By analyzing the recovery time and decay pattern, the response inertia of the internal electrical structure can be deduced. Different types of latent defects, such as grid line breakage or encapsulation layer damping, will exhibit specific nonlinear characteristics during the decay phase, making fault identification more targeted.

[0064] To address the uncertainty of signals under random disturbances, a time-stamped sequence can be introduced at the data acquisition end, marking the start and end times of each disturbance event in the response data stream. During subsequent processing, these time stamps serve as segmentation points, dividing the electrical parameter data into several independent events, which are then normalized and compensated for separately. This not only avoids the superposition of consecutive disturbances but also generates results from multiple independent experiments in signal statistics, significantly improving the robustness of the analysis.

[0065] In harsh power plant environments, such as high-latitude winters with strong winds and low temperatures, this random pulse envelope method exhibits higher robustness than traditional sinusoidal perturbations. Because it does not rely on the sustained stability of periodic signals, even short-term fluctuations in irradiance during testing will not significantly deviate from the results. Furthermore, this method generates a wider dynamic signal amplitude and a richer spectral energy distribution, covering more frequency bands of electrical characteristics, thus obtaining more comprehensive response information in a single test.

[0066] Furthermore, in situations where communication link bandwidth is limited or inverter control refresh rate is low in some photovoltaic power plants, this solution can achieve intermittent excitation through discrete pulse mode, requiring only sampling of the response during the intervals, without the need for continuous communication control. This allows dynamic detection to be performed independently at edge nodes or in low-bandwidth scenarios, making it suitable for unattended or off-grid photovoltaic power plants.

[0067] Step S30: Determine the impedance spectrum characteristics and extract the dynamic response fingerprint based on the dynamic response data, and construct a feature vector to characterize the string electrical state.

[0068] Specifically, a Fast Fourier Transform is performed on the voltage response sequence and current response sequence after operating condition compensation to calculate the complex impedance spectrum, and the corresponding impedance spectrum feature parameters are extracted based on the complex impedance spectrum; wherein, the impedance spectrum feature parameters include any one or more of impedance magnitude, impedance phase angle, equivalent series resistance, and parasitic capacitance; dynamic fingerprint features are extracted in the frequency domain and time domain of the dynamic response data respectively; wherein, the dynamic fingerprint features include any one or more of main harmonic amplitude, phase delay, harmonic energy ratio, response time constant, and nonlinear exponent; the impedance spectrum feature parameters and dynamic fingerprint features are concatenated in a preset order to form a feature vector, which is used as a feature vector to characterize the string electrical state.

[0069] Furthermore, the impedance spectrum feature parameters and dynamic fingerprint features are concatenated into a feature vector in a preset order to characterize the string electrical state. This includes: sorting the impedance spectrum feature parameters and dynamic fingerprint features according to their sensitivity to the string electrical state, generating a feature index table to indicate the concatenation position; performing normalization and dimension unification processing on the impedance spectrum feature parameters and dynamic fingerprint features respectively, and writing them sequentially into the same data structure according to the order of the feature index table to form a feature vector characterizing the string electrical state.

[0070] In this embodiment of the invention, after completing the acquisition of dynamic response data and operating condition compensation, a more diagnostically valuable step is needed: analyzing the dynamic signals to extract characteristic quantities that reflect the true electrical health state of the string. This step is the core of the entire method because by analyzing the frequency and time domain characteristics of the voltage and current responses, changes in the internal electrical characteristics of the string can be revealed, such as increased contact resistance, attenuation of capacitance characteristics, or nonlinear shifts in the current conduction path, which are often completely masked in steady-state measurements.

[0071] In the specific implementation process, the voltage response sequence and current response sequence after operating condition compensation are first input into the Fast Fourier Transform (FFT) module for synchronous frequency domain transformation. After obtaining the voltage and current spectra, the two are compared to obtain the complex impedance spectrum Z(f), which contains the amplitude and phase information of the impedance. For an ideal and healthy photovoltaic string, the impedance spectrum should exhibit a regular single-peak distribution with smooth phase angle changes; however, when poor contact, encapsulation humidification, or bypass diode deterioration occurs, the impedance spectrum will exhibit characteristics such as arc contraction, tilting, or even discontinuities.

[0072] When extracting impedance spectrum features, parameters such as impedance magnitude |Z(f)|, impedance phase angle θ(f), equivalent series resistance Rs, and parasitic capacitance Cp can be calculated. Rs mainly reflects the integrity of the conductive path of the string; an increase in Rs often indicates increased contact resistance or solder joint aging. Cp corresponds to the change in parasitic capacitance between components; Cp will decrease when the encapsulation layer absorbs moisture or the insulation between the electrodes degrades. By analyzing the variation of the impedance spectrum in different frequency bands, it is also possible to distinguish between large-area power attenuation and local connection problems. For example, anomalies in the low-frequency range are usually related to material aging or electrode damage, while anomalies in the high-frequency range are more related to cable shielding or solder joint problems.

[0073] Simultaneously, beyond the frequency domain, the time-domain characteristics of the response signal must also be analyzed, which is the so-called dynamic fingerprint. The dynamic fingerprint is not a single parameter, but a set of indicators characterizing dynamic behavior. Among them, the amplitude of the main harmonic reflects the overall energy transfer capability of the response, and the phase delay describes the time lag of the current relative to the voltage; this indicator is particularly sensitive to identifying electrical nonlinearity. When components exhibit capacitive aging or internal potential traps, the response phase will show a significant shift. The harmonic energy ratio, obtained by comparing the energy of the main frequency with that of higher-order harmonics, is used to reflect the degree of distortion in the response waveform. Typically, the harmonic energy ratio of a healthy string is stable in a low range, while the proportion of high-frequency harmonic energy will increase significantly when there are diode abnormalities or partial short circuits.

[0074] Furthermore, the response time constant and nonlinear exponent can be extracted. The time constant describes the speed at which the system returns to steady state after the disturbance is removed, and it comprehensively reflects the discharge characteristics of energy storage elements (such as junction capacitance and parasitic inductance). The nonlinear exponent describes the deviation of the current-voltage response curve, and can be obtained by fitting the residual of the current-voltage curve. This exponent typically increases when the components are in the aging stage. Through the combined analysis of these dynamic indicators, an accurate assessment of the string health status can be obtained without disassembly or interruption of operation.

[0075] After extracting these frequency and time domain parameters, they need to be combined into a unified structured vector, also known as an eigenvector. To ensure comparability between different physical quantities, all parameters must first be normalized, and the influence between different units must be eliminated through a dimensionless method. For example, parameters of different orders of magnitude, such as impedance magnitude, phase angle, and power ratio, can only be treated with equal weight after dimensionless standardization.

[0076] The order in which the eigenvectors are concatenated is not arbitrary, but rather ordered according to the sensitivity of each parameter to the electrical state. Generally, changes in impedance magnitude and phase angle reflect faults first, and are therefore listed first; dynamic features such as phase delay and harmonic energy ratio are next; time constant and nonlinear exponent are used to supplement the judgment and are usually placed at the end of the vector. To this end, a feature index table is first generated, assigning a fixed index number and weight coefficient to each feature, and then writing them sequentially into the data structure. Each feature dimension corresponds to a clear physical meaning, making the eigenvectors not only usable for algorithmic identification but also interpretable.

[0077] To ensure model consistency over long-term operation, timestamps or sampling batch numbers can be added during feature concatenation to identify the source period of the feature vectors. This is crucial for subsequent health template comparison, as environmental differences at different times can cause slight shifts, and time stamps can help eliminate the effects of seasonality or diurnal variations.

[0078] Furthermore, to further enhance the discriminative power of eigenvectors, certain features can be derived and combined. For example, the product of the impedance phase angle and the harmonic energy ratio can be used as an energy-phase synergy index to reveal the coupling relationship between the phase and energy distortion of the response. As another example, the ratio of the main harmonic amplitude to the nonlinearity index can be used as a dynamic sensitivity coefficient to determine the trend of string response amplitude changes under the same disturbance amplitude. Although these derived features are secondary indicators, they often provide more stable diagnostic criteria in practical judgments.

[0079] After constructing the feature vectors, they can be directly used to characterize the electrical state of the strings, which is a compressed description of health features. The feature vectors of healthy strings will cluster in a certain region in multidimensional space, while strings with abnormalities will deviate significantly from this cluster center.

[0080] This stage of processing transforms dynamic response signals into quantifiable electrical health characteristics. Unlike traditional analyses that focus solely on power changes or current fluctuations, this approach not only considers the amplitude of electrical parameters but also captures deeper information such as phase, spectral energy distribution, and nonlinear evolution. This elevates the detection method from macroscopic indicator monitoring to electrical behavior-level analysis. In other words, it can not only identify which string is abnormal but also infer the physical mechanism of the abnormality, such as contact aging, capacitor decay, or diode reverse leakage.

[0081] Step S40: Compare the feature vector with the health template and perform phase coupling analysis to generate fault determination results and output the corresponding string location information.

[0082] Specifically, the system reads the health template vector corresponding to the target string from the historical operation archive, performs distance metric calculation on the current feature vector and the health template, and obtains the deviation index of each feature component; calculates the response phase correlation coefficient between the feature vectors of each string within the same bus unit range, and constructs a phase coupling matrix; when the deviation of any string exceeds a preset threshold, and its correlation in the phase coupling matrix deviates significantly from the health template, the string is identified as an abnormal object; based on the power station wiring log, the abnormal object is mapped to the actual physical location, and the location result including string number, fault type and confidence level is output.

[0083] Furthermore, within the same bus unit, the response phase correlation coefficient between the feature vectors of each string is calculated to construct a phase coupling matrix. This includes: extracting the corresponding phase delay feature components from the feature vectors of each string, and calculating the phase change sequence of adjacent sampling periods within a preset time window; performing correlation coefficient calculation on the phase change sequences of any two strings to obtain phase correlation parameters, and arranging all phase correlation parameters according to the string number to generate a phase coupling matrix.

[0084] In this embodiment of the invention, after obtaining the dynamic feature vectors of each string, the next task is to determine the degree of difference between these feature vectors and the ideal healthy state, which is the so-called health deviation analysis. This step is both the basis for diagnosis and the core basis for fault location. Simply put, it answers two questions: Is the string "abnormal," and to what extent is it abnormal? And does the abnormality have an electrical coupling relationship with other strings? To achieve this, two types of calculations need to be performed simultaneously: one is "individual deviation," and the other is "phase coupling." The two corroborate each other to ultimately provide an accurate fault determination.

[0085] First, a health template needs to be established, which is a record of the feature vectors of each string under fault-free and normal operating conditions. This data can come from the calibration phase during the initial commissioning of the power plant, or it can be formed from the feature statistics of a long-term stable operation phase. The health template typically contains multi-dimensional parameters, including impedance characteristics, phase characteristics, and the average behavior of dynamic fingerprints. Each parameter dimension records a reference interval or mean vector for subsequent comparison.

[0086] In practice, the currently detected feature vector is compared dimension-by-dimensionally with the corresponding template vector. This comparison is not a simple numerical subtraction, but rather a calculation of distance metrics, commonly weighted Euclidean distance or cosine similarity. The former measures the overall deviation magnitude, while the latter measures the direction of change. Distance calculations yield a deviation index for each feature component. For example, changes in impedance phase angle have a significant impact on electrical coupling and can be assigned a higher weight; while power-related parameters have relatively lower weights. This deviation index not only tells us "this string is not right," but also indicates which type of parameter has changed significantly.

[0087] The next step is phase coupling analysis, which aims to identify whether the anomaly is "independent" or "driven." Within the same bus unit, the phase delay component is extracted from the feature vectors of all strings, and a phase change record is established as a time sequence. Within a predefined time window (e.g., 5 or 10 minutes), the correlation coefficient between the phase change sequences of any two strings is calculated. This correlation coefficient characterizes the similarity of their responses. If the two curves are highly synchronized in time, it indicates that their electrical behavior is affected by the same disturbance; conversely, if the phase behavior of a string suddenly deviates from the overall pattern, it means that there may be a local fault. The correlation coefficients between all strings are arranged into a matrix to obtain the phase coupling matrix.

[0088] The diagonal elements of this matrix are 1, representing autocorrelation, while the values ​​of the remaining elements range from 0 to 1. Under healthy conditions, the correlation coefficient between adjacent strings should be high, while the correlation between non-adjacent strings decreases with spatial distance. If a string shows a significantly decreased correlation with its physically neighboring strings, or conversely, an abnormally high correlation with distant strings, it indicates a possible abnormality in its electrical connection, such as incorrect cross-string wiring, local grounding, or insulation damage.

[0089] After calculating the deviation index and the phase coupling matrix, the two are used together for judgment. When the characteristic deviation of a string exceeds a preset threshold, and its correlation pattern in the phase coupling matrix is ​​significantly different from the healthy template, it is judged as an abnormal string. The "dual-condition triggering" here can effectively avoid false alarms. For example, when the overall irradiance fluctuation causes all strings to show a slight deviation synchronously, the phase coupling analysis will show a high degree of consistency, and the system will not mistakenly judge it as a fault. Conversely, if only the phase correlation of one or a few strings is disrupted, it can be basically confirmed that it is a local problem.

[0090] To make the results more actionable, the serial numbers of the abnormal fault groups are mapped one-to-one with their physical locations in the power plant wiring log. This allows for direct location of the specific array, combiner box, or even serial number during the output phase. The output includes not only the fault group number and fault type but also a confidence score. The confidence score is calculated using the joint distribution of deviation and correlation coefficient; a higher score indicates a more certain anomaly. For example, a fault with high deviation and a significantly reduced correlation coefficient typically has a confidence score above 90%; conversely, a slight deviation results in a lower confidence score.

[0091] In another implementation, a time-series sliding window strategy can be introduced based on the construction of the phase coupling matrix. Specifically, instead of calculating the correlation within a fixed time period, the window scrolls along the time axis, updating only a subset of data points each time. This allows for more sensitive detection of sudden coupling changes. For intermittent faults, such as intermittent grounding caused by moisture, traditional static matrices often fail to respond promptly, while the dynamic updates of the sliding window allow anomalies to be identified within seconds. Furthermore, cluster analysis of the matrix can automatically group strings with similar electrical responses into the same category, while strings not belonging to any cluster are directly marked as isolated points, providing another dimension to assist in fault identification.

[0092] By simultaneously referencing static differences (feature deviation) and dynamic correlations (phase correlation), the one-sidedness of judging by a single indicator is avoided, and local anomalies hidden in group behavior can be identified more reliably.

[0093] like Figure 2 As shown, this invention provides a rapid fault location system for photovoltaic power plants. The system includes: a data acquisition unit, used to acquire electrical parameter data of each string in the photovoltaic power plant and perform normalization processing in conjunction with environmental conditions to generate baseline data, and to screen suspicious strings based on the baseline data; a feedback unit, used to apply a preset power perturbation signal to the suspicious strings from the inverter side within a preset amplitude and preset frequency range, and to synchronously acquire the corresponding instantaneous voltage and current values ​​during the duration of the perturbation, and to form dynamic response data after detrending processing and operating condition compensation; a processing unit, used to determine impedance spectrum characteristics based on the dynamic response data and extract dynamic response fingerprints to construct feature vectors for characterizing the electrical state of the strings; and an output unit, used to compare the feature vectors with a health template and perform phase coupling analysis to generate fault judgment results and output corresponding string location information.

[0094] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0095] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0096] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for fast fault location of a photovoltaic power plant, characterized in that, The method comprises: Obtaining the electrical parameter data of each group of strings of the photovoltaic power station and performing normalization processing combined with the environmental working conditions to generate baseline data, and screening suspicious groups of strings based on the baseline data; wherein, Obtaining the electrical parameter data of each group of strings of the photovoltaic power station and performing normalization processing combined with the environmental working conditions to generate baseline data, comprising: collecting the voltage signal and current signal of each group of strings within a preset sampling period, and the corresponding irradiance, environmental temperature and wind speed parameters, and constructing a sampling set containing multi-dimensional working condition parameters; calculating the working condition correction coefficient of each group of strings according to the sampling set, and performing normalization operation on the voltage and current data in the sampling set to form baseline data reflecting the output characteristics of each group of strings under the same working conditions; The rule for screening suspicious groups of strings based on the baseline data is: calculating three types of normalized indexes of voltage deviation, current deviation and power stability in the baseline data of each group of strings, and performing sliding window statistical analysis on the normalized indexes within a continuous sampling period; when the average absolute value of the voltage deviation or the current deviation of any group of strings exceeds the preset threshold value, and the power stability is lower than the lower limit of the mean value of the adjacent groups of strings, or there is a trend deviation in the adjacent time windows, the group of strings is determined as a suspicious group of strings; Applying a preset power perturbation signal within a preset amplitude and a preset frequency range to the suspicious group of strings from the inverter side, and synchronously collecting the corresponding voltage instantaneous value and current instantaneous value during the perturbation duration, forming dynamic response data after detrending and working condition compensation; comprising: The maximum power point tracking channel of the inverter side applies a periodic perturbation to the power reference value of the target group of strings within a preset amplitude and a preset frequency range, and records the corresponding perturbation control signal; synchronously sampling the voltage and current instantaneous values of the target group of strings during the perturbation duration, forming a voltage response sequence and a current response sequence; performing detrending and working condition compensation on the voltage response sequence and the current response sequence to eliminate the common mode influence of irradiance and temperature, and aligning with the perturbation control signal on the time axis to generate corrected dynamic response data; Performing detrending and working condition compensation on the voltage response sequence and the current response sequence to eliminate the common mode influence of irradiance and temperature, comprising: extracting the corresponding irradiance variation curve and environmental temperature variation curve during the perturbation duration, and calculating the instantaneous slope coefficient of irradiance to current and the correction coefficient of temperature to voltage; constructing a working condition compensation model according to the slope coefficient and the correction coefficient, and performing linear detrending and differential normalization processing on the voltage response sequence and the current response sequence, respectively; Based on the dynamic response data, determine the impedance spectrum characteristics and extract the dynamic response fingerprint to construct a feature vector for characterizing the electrical state of the group of strings; comprising: performing fast Fourier transform on the voltage response sequence and the current response sequence after the working condition compensation to calculate a complex impedance spectrum, and extracting corresponding impedance spectrum feature parameters based on the complex impedance spectrum; wherein the impedance spectrum feature parameters include any one or more of impedance modulus, impedance phase angle, equivalent series resistance, and parasitic capacitance; extracting dynamic fingerprint features in the frequency domain and the time domain of the dynamic response data respectively; wherein the dynamic fingerprint features include any one or more of main harmonic amplitude, phase delay, harmonic energy ratio, response time constant, and nonlinearity index; splicing the impedance spectrum feature parameters and the dynamic fingerprint features into a feature vector in a preset order as a feature vector for representing the electrical state of the string; comparing the feature vector with a health template and performing phase coupling analysis to generate a fault determination result and output corresponding string positioning information.

2. The method for fast locating fault of a photovoltaic power station according to claim 1, characterized in that, splicing the impedance spectrum feature parameters and the dynamic fingerprint features into a feature vector in a preset order as a feature vector for representing the electrical state of the string, including: sorting the sensitivity of each impedance spectrum feature parameter and each dynamic fingerprint feature to the electrical state of the string to generate a feature index table for indicating the splicing position; performing normalization and dimension unification processing on the impedance spectrum feature parameters and the dynamic fingerprint features respectively, and sequentially writing into the same data structure according to the order of the feature index table to form a feature vector for representing the electrical state of the string.

3. The method for fast locating fault of photovoltaic power station according to claim 1, characterized in that, comparing the feature vector with a health template and performing phase coupling analysis to generate a fault determination result and output corresponding string positioning information, including: reading a health template vector corresponding to the target string from the historical operation archives, performing distance measurement calculation on the current feature vector and the health template to obtain the deviation index of each feature component; calculating the response phase correlation coefficient between the feature vectors of each string within the same busbar unit to construct a phase coupling matrix; when the deviation of any string exceeds a preset threshold, and the correlation in the phase coupling matrix deviates significantly from the health template, the string is determined as an abnormal object; mapping the abnormal object to the actual physical location according to the power station wiring account to output the positioning result including the string number, fault type, and confidence.

4. The method for fast locating fault of a photovoltaic power station according to claim 3, characterized in that, calculating the response phase correlation coefficient between the feature vectors of each string within the same busbar unit to construct a phase coupling matrix, including: extracting the phase delay feature component corresponding to each string feature vector, and calculating the phase change sequence of adjacent sampling periods within a preset time window; performing correlation coefficient calculation on the phase change sequences of any two strings to obtain the phase correlation parameter, and arranging all the phase correlation parameters according to the string number to generate a phase coupling matrix.

5. A photovoltaic power plant fault fast location system characterized by, The system is used to perform the photovoltaic power station fault rapid positioning method of any one of claims 1-4, and the system includes: an acquisition unit configured to acquire electrical parameter data of each string of the photovoltaic power station, perform normalization processing in combination with the environmental working condition, generate baseline data, and filter suspicious strings based on the baseline data; A feedback unit is configured to apply a preset power perturbation signal within a preset amplitude and frequency range on the suspicious string by an inverter side, and synchronously collect corresponding voltage instantaneous values and current instantaneous values during the perturbation duration, and form dynamic response data after de-trending processing and working condition compensation; A processing unit is configured to determine impedance spectrum features and extract dynamic response fingerprints based on the dynamic response data, and construct a feature vector for representing an electrical state of the string; An output unit is configured to compare the feature vector with a health template, perform phase coupling analysis, generate a fault determination result, and output corresponding string positioning information.

Images