Floating type photovoltaic fault time sequence feature adaptive extraction diagnosis method and device

Abstract

The invention discloses a floating type photovoltaic fault time sequence characteristic self-adaptive extraction diagnosis method and device, and the method comprises the steps: obtaining floating body time domain inclination angle data through a floating body reduced scale model test, and constructing an offshore photovoltaic array power generation characteristic model in combination with a photovoltaic cell engineering model to generate a fault I-V data curve; carrying out characteristic analysis on the fault I-V data curve, and selecting a fault characteristic quantity; carrying out clustering analysis through a fuzzy C-means clustering algorithm, and determining a historical moment similar to the wave influence at the current moment; the cosine similarity of converted I-V curve feature vectors is obtained through calculation, the evolution degree of a fault time sequence is judged to determine the width of a self-adaptive window, and an effective fault feature sequence is extracted; and a CBRM deep network model is constructed and trained, diagnosis processing is carried out through the trained model, and an offshore photovoltaic array fault diagnosis result is obtained. The method solves the problems that in the prior art, misjudgment and missed judgment are likely to happen, and a fixed time window cannot adapt to non-stationary feature changes.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic fault diagnosis technology, specifically to an adaptive extraction and diagnosis method and apparatus for the temporal characteristics of floating photovoltaic faults. Background Technology

[0002] The large-scale development of renewable energy is accelerating. Offshore floating photovoltaic (PV) systems, with their unique advantages such as not occupying land resources, excellent component heat dissipation, and reduced water surface evaporation, have become an important development direction in the PV field, and are gradually being demonstrated and deployed on a large scale in nearshore waters and lakes. Currently, fault diagnosis of offshore PV arrays mainly follows the technical path of onshore PV systems, relying primarily on diagnostic methods based on IV curve feature analysis. This involves periodically inspecting and collecting current-voltage data from the PV array to extract features such as open-circuit voltage, short-circuit current, and maximum power point parameters, and then combining these with preset thresholds or traditional machine learning models to determine the fault type. Meanwhile, some studies are attempting to introduce time-series feature analysis, using fixed time windows to extract feature sequences to improve the ability to identify gradual-type faults. These technologies have been widely applied and have achieved certain results in stable onshore environments.

[0003] However, the operating environment of offshore photovoltaic arrays differs fundamentally from that on land, making it difficult for existing technologies to adapt to their complex operating conditions, resulting in low fault diagnosis accuracy. On one hand, wind and wave disturbances at sea cause changes in the attitude of the floating structure, uneven irradiance on the component surface, and fluctuations in cable contact conditions, causing the IV characteristic curve of the photovoltaic array to exhibit a multi-knee phenomenon with multiple inflection points. This phenomenon is highly similar to real fault characteristics, leading traditional IV curve-based diagnostic methods to easily misjudge wave disturbances as faults, or miss real faults due to feature aliasing, resulting in high false positive and false negative rates. On the other hand, existing diagnostic methods that incorporate time-series features employ a fixed time window strategy, which cannot adaptively cope with the non-stationary changes in fault characteristics under wave disturbances, making it difficult to accurately capture the evolution of faults over time. Furthermore, traditional methods do not effectively mathematically model wave dynamic disturbances and lack targeted robust diagnostic strategies, failing to fundamentally distinguish the essential differences between wave disturbances and real faults, severely restricting the safe and stable operation of offshore photovoltaic systems.

[0004] Therefore, there is an urgent need for an adaptive extraction and diagnosis method for the time-series features of floating photovoltaic faults to solve the problems of easy misjudgment and missed judgment in existing technologies, and the inability of fixed time windows to adapt to non-stationary feature changes. Summary of the Invention

[0005] To address this, the present invention provides an adaptive extraction and diagnosis method and apparatus for the temporal features of floating photovoltaic faults, which solves the problems of easy misjudgment and missed judgment in the prior art, and the inability of fixed time windows to adapt to non-stationary feature changes, thereby achieving accurate and efficient online fault diagnosis.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults, comprising: Through scaled-down model tests of the floating body, time-domain tilt angle data of the floating body under the action of sea waves were obtained; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, a marine photovoltaic array power generation characteristic model was constructed; through the marine photovoltaic array power generation characteristic model, fault IV data curves under set wave conditions and fault types were generated. The fault IV data curve was subjected to characteristic analysis, and the open circuit voltage, short circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate were selected as fault characteristic quantities. Based on the floating body tilt angle volatility, the historical volatility data is clustered using the fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact. The fault IV data at the current time and similar historical time are converted according to standard test conditions, and the cosine similarity of the feature vector of the converted IV curve is calculated. The degree of evolution of the fault time sequence is judged based on the cosine similarity, and the judgment result is obtained. Based on the judgment result, the adaptive window width containing complete fault evolution information is determined, and the effective fault feature sequence is extracted. A deep network model for CBRM is constructed by fusing CNN, residual connection BiGRU, and MHA; the effective fault feature sequence is input into the deep network model for training and iterative optimization to obtain a trained deep network model; the trained deep network model is then used for diagnostic processing to obtain fault diagnosis results for marine photovoltaic arrays.

[0007] As a preferred embodiment of an adaptive extraction and diagnosis method for the timing characteristics of floating photovoltaic faults, the expression for the power generation characteristic model of the offshore photovoltaic array is as follows: ; In the formula, For the output current of the floating photovoltaic array; The number of parallel strings; This represents the number of components connected in series. This is the short-circuit current; , All are parameters; This refers to the output voltage of the floating photovoltaic array. This is the open-circuit voltage; , These are the differentials of current and voltage, respectively.

[0008] In a preferred embodiment of an adaptive extraction and diagnostic method for the timing characteristics of floating photovoltaic system faults, the floating body tilt angle fluctuation rate is measured by a gyroscope device monitoring the floating body tilt angle and calculated by a monitoring host; the expression for the floating body tilt angle fluctuation rate is: ; In the formula, The undulation rate of the floating body's tilt angle; Let be the tilt angle of the component at the i-th sampling point; This represents the number of sampling points within that time period. This represents the average tilt angle of the photovoltaic module.

[0009] As a preferred scheme for an adaptive extraction and diagnosis method of fault timing features in floating photovoltaic systems, the conversion formula is as follows when converting fault IV data from the current moment to similar historical moments under standard test conditions: ; ; In the formula, The current under STC conditions; The voltage under STC conditions; and These are the raw current and voltage data collected from the photovoltaic array, respectively. , These are the reference irradiance and reference temperature under standard test conditions, respectively; , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively; G is the irradiance received by the tilted surface; and T is the original photovoltaic array temperature. This is the open-circuit voltage irradiance coefficient; The formula for calculating the cosine similarity of the eigenvectors of the IV curves after conversion is as follows: ; In the formula, Cosine similarity; The eigenvectors of the IV curve under STC conditions; This represents the eigenvectors of IV curves at similar historical moments.

[0010] As a preferred embodiment of the adaptive extraction and diagnosis method for the time-series features of floating photovoltaic faults, in the process of determining the adaptive window width containing complete fault evolution information based on the judgment result, if the adaptive window width is greater than the set maximum window width, or if there are fewer than 3 similar historical sampling points, it is determined to be a strongly non-stationary state, and diagnosis is temporarily not performed.

[0011] This invention also provides an adaptive extraction and diagnosis device for the timing features of floating photovoltaic faults, based on the above-mentioned adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults, comprising: The Fault IV data curve generation module is used to obtain the time-domain tilt angle data of the floating body under the action of sea waves through the floating body scale model test; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, a marine photovoltaic array power generation characteristic model is constructed; and the Fault IV data curve is generated under the set wave conditions and fault type through the marine photovoltaic array power generation characteristic model. The Fault IV data curve characteristic analysis module is used to perform characteristic analysis on the Fault IV data curve, and selects open circuit voltage, short circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate as fault characteristic quantities. The similar historical moment determination module is used to perform cluster analysis on historical volatility data based on the floating body tilt angle volatility using a fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact. The effective fault feature sequence extraction module is used to convert fault IV data at the current time and similar historical time according to standard test conditions, calculate the cosine similarity of the feature vector of the converted IV curve; judge the degree of evolution of the fault time sequence based on the cosine similarity, and obtain the judgment result; based on the judgment result, determine the adaptive window width containing complete fault evolution information, and extract the effective fault feature sequence. The CBRM deep network model construction and processing module is used to construct a CBRM deep network model by fusing CNN, residual connection BiGRU and MHA; input the effective fault feature sequence into the CBRM deep network model for training and iterative optimization to obtain a trained CBRM deep network model; and perform diagnostic processing through the trained CBRM deep network model to obtain fault diagnosis results for marine photovoltaic arrays.

[0012] As a preferred embodiment of an adaptive extraction and diagnostic device for the timing characteristics of floating photovoltaic faults, the expression for the power generation characteristic model of the offshore photovoltaic array in the fault IV data curve generation module is as follows: ; In the formula, For the output current of the floating photovoltaic array; The number of parallel strings; This represents the number of components connected in series. This is the short-circuit current; , All are parameters; This refers to the output voltage of the floating photovoltaic array. This is the open-circuit voltage; , These are the differentials of current and voltage, respectively.

[0013] As a preferred embodiment of an adaptive extraction and diagnostic device for the timing characteristics of floating photovoltaic faults, in the fault IV data curve characteristic analysis module, the floating body tilt angle fluctuation rate is tested by a gyroscope device monitoring the floating body tilt angle and calculated by the monitoring host; the expression for the floating body tilt angle fluctuation rate is: ; In the formula, The undulation rate of the floating body's tilt angle; Let be the tilt angle of the component at the i-th sampling point; This represents the number of sampling points within that time period. This represents the average tilt angle of the photovoltaic module.

[0014] As a preferred embodiment of a floating photovoltaic fault timing feature adaptive extraction and diagnostic device, in the effective fault feature sequence extraction module, during the process of converting fault IV data from the current moment to similar historical moments according to standard test conditions, the conversion formula is as follows: ; ; In the formula, The current under STC conditions; The voltage under STC conditions; and These are the raw current and voltage data collected from the photovoltaic array, respectively. , These are the reference irradiance and reference temperature under standard test conditions, respectively; , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively; G is the irradiance received by the tilted surface; and T is the original photovoltaic array temperature. This is the open-circuit voltage irradiance coefficient; The formula for calculating the cosine similarity of the eigenvectors of the IV curves after conversion is as follows: ; In the formula, Cosine similarity; The eigenvectors of the IV curve under STC conditions; This represents the eigenvectors of IV curves at similar historical moments.

[0015] As a preferred embodiment of a floating photovoltaic fault time-series feature adaptive extraction and diagnostic device, in the effective fault feature sequence extraction module, during the process of determining the adaptive window width containing complete fault evolution information based on the judgment result, if the adaptive window width is greater than the set maximum window width, or the number of similar historical time sampling points is less than 3, it is determined to be a strongly non-stationary state, and diagnosis is temporarily not performed.

[0016] This invention has the following advantages: It obtains time-domain tilt angle data of a floating body under the action of ocean waves through a scaled-down model test; based on this time-domain tilt angle data, and combined with a photovoltaic cell engineering model, it constructs a power generation characteristic model of an offshore photovoltaic array; through this model, it generates fault IV data curves under set wave conditions and fault types; it performs characteristic analysis on the fault IV data curves, selecting open-circuit voltage, short-circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current, and floating body tilt angle volatility as fault characteristic quantities; based on the floating body tilt angle volatility, it performs cluster analysis on historical volatility data using a fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact. The invention involves: 1) Reconverting fault IV data from the current time point to similar historical time points under standard test conditions, calculating the cosine similarity of the feature vectors of the converted IV curves; 2) Judging the degree of fault time-series evolution based on the cosine similarity, obtaining a judgment result; 3) Determining the adaptive window width containing complete fault evolution information based on the judgment result, and extracting effective fault feature sequences; 4) Constructing a CBRM deep network model by fusing CNN, residual connection BiGRU, and MHA; 5) Inputting the effective fault feature sequences into the CBRM deep network model for training and iterative optimization, obtaining a trained CBRM deep network model; 6) Performing diagnostic processing through the trained CBRM deep network model to obtain fault diagnosis results for the marine photovoltaic array. Compared to theoretical models based on Stokes waves, this invention uses wave data obtained from scaled model experiments that is more accurate, and the established marine photovoltaic array model can simulate more realistic wave conditions. This invention proposes an adaptive time window feature extraction strategy, which can determine the length of feature sequences that fully reflect fault evolution under wave interference. Compared to a fixed time window, this method has higher accuracy under different wave environments. This invention designs a temporal fusion network model CBRM that integrates a CNN model, a BiGRU model with residual connections, and an MHA mechanism. This model can suppress interference from wave environments and achieve more accurate fault diagnosis. This invention can be directly deployed in offshore photovoltaic power plants for routine fault diagnosis and early warning, demonstrating high engineering application value. Attached Figure Description

[0017] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0018] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0019] Figure 1 This is a flowchart illustrating an adaptive extraction and diagnosis method for timing features of floating photovoltaic faults provided in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the modeling and fault feature analysis process of the power generation characteristics of a marine photovoltaic array in an adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults provided in Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the fault features and temporal evolution of a photovoltaic array under wave influence in an adaptive extraction and diagnosis method for the temporal features of floating photovoltaic faults provided in Embodiment 1 of the present invention. Figure 4 This is a schematic diagram of the process for determining the similarity of wave influence based on fuzzy C-means clustering in the adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults provided in Embodiment 1 of the present invention. Figure 5 This is a schematic diagram of the adaptive time window feature sequence extraction process in the adaptive extraction and diagnosis method for time-series features of floating photovoltaic faults provided in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the CBRM network model processing flow in the adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults provided in Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of a simulation model of a marine photovoltaic array in one possible embodiment provided in Embodiment 1 of the present invention; Figure 8 This is a schematic diagram of the adaptive fault diagnosis results of marine photovoltaic systems in one possible embodiment provided in Embodiment 1 of the present invention; Figure 9 This is a schematic diagram of the architecture of a floating photovoltaic fault timing feature adaptive extraction and diagnosis device provided in Embodiment 2 of the present invention. Detailed Implementation

[0020] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1

[0022] See Figure 1 Embodiment 1 of the present invention provides an adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults, comprising the following steps: S1. Obtain the time-domain tilt angle data of the floating body under the action of sea waves through a scaled-down model test of the floating body; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, construct a marine photovoltaic array power generation characteristic model; generate fault IV data curves under set wave conditions and fault types through the marine photovoltaic array power generation characteristic model. S2. Perform characteristic analysis on the fault IV data curve and select open circuit voltage, short circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate as fault characteristic quantities. S3. Based on the floating body tilt angle volatility, perform cluster analysis on the historical volatility data using the fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact; S4. Convert the fault IV data at the current time and similar historical time according to standard test conditions, and calculate the cosine similarity of the feature vector of the converted IV curve; judge the degree of evolution of the fault time sequence based on the cosine similarity, and obtain the judgment result; based on the judgment result, determine the adaptive window width containing complete fault evolution information, and extract the effective fault feature sequence. S5. By fusing CNN, residual connection BiGRU, and MHA, a CBRM deep network model is constructed; the effective fault feature sequence is input into the CBRM deep network model for training and iterative optimization to obtain a trained CBRM deep network model; the trained CBRM deep network model is used for diagnostic processing to obtain fault diagnosis results for the marine photovoltaic array.

[0023] In this embodiment, in step S1, the time-domain tilt angle data of the floating body under the action of sea waves is obtained through a scaled-down model test of the floating body; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, a marine photovoltaic array power generation characteristic model is constructed; through the marine photovoltaic array power generation characteristic model, fault IV data curves under set wave conditions and fault types are generated.

[0024] Specifically, such as Figure 2 As shown, in the field of marine engineering, based on the hydrological parameters of nearshore waters, ocean waves are usually irregular waves, and their spectral distribution can be described by the JONSWAP spectrum, as shown in equation (1): (1) In the formula, JONSWAP spectrum of wavefront height; For the reason The constant that determines the outcome; For the sake of righteousness, the waves rise high; The periodicity of the spectral peak; For frequency components; For shape parameters; For peak shape parameters.

[0025] In this embodiment, a scaled-down model test in a water tank is required to test the motion response of the floating body under real waves. The simulation of wave conditions, floating structure, and mooring system is typically based on the Froude criterion. Wave motion is a stationary, ergodic stochastic process. Waves are generated in the laboratory using a wave generator, and their displacement spectrum can be calculated using the following equation (2): (2) In the formula, the angular frequency ; The displacement spectrum of the wave-generating plate; This is a transfer function.

[0026] The transfer function of the wave generator and the wave is calculated by the following equation (3): (3) In the formula, For water depth; For wave number.

[0027] A scaled-down model of the floating platform was constructed in a controlled environment to simulate real finite-amplitude wave conditions. The motion response of the floating body's tilt angle was tested under different structural conditions, and the time-domain tilt angle data of the floating body was obtained by measuring it in real time using equipment such as gyroscopes. .

[0028] To obtain the photovoltaic array power generation characteristics under wave conditions, it is necessary to consider the irradiance received by the modules on the floating structure. Here, some necessary simplifications are made to this modeling process: only the pitch motion of the floating body caused by waves is considered, and the horizontal motion is ignored; it is assumed that the wave propagation occurs in a single direction; the change in the tilt angle of the floating body caused by waves is equal to the change in the tilt angle of the photovoltaic modules; the motion of the floating structure under wave action conforms to nonlinear wave theory.

[0029] For a common rigid floating block structure, photovoltaic modules are mounted on its surface. The tilt angle of the module relative to the horizontal plane in the wave direction is expressed as... The irradiance received by the inclined surface can be calculated by equation (4): (4) In the formula, This refers to the direct irradiance on a horizontal plane. This refers to the horizontal diffuse irradiance. It is the horizontal diffuse irradiance. , The initial tilt angle set for the photovoltaic modules. The angle between sunlight and the normal to the tilted photovoltaic modules. It can be calculated using the following formula (5): (5) In the formula, It is the declination angle; Latitude; The hour angle; It is the azimuth angle.

[0030] The time-domain tilt angle of the floating body obtained from the scaled-down model test Substituting into equation (4), the surface irradiance of the offshore photovoltaic module can be calculated: (6) To apply the established model to practical engineering applications, this invention employs an engineering model of photovoltaic cells for further modeling. This model is described using only four known parameters specified at the time of manufacture of the photovoltaic module, including the maximum power point voltage. Current Open circuit voltage Short circuit current As described in equation (7): (7) In the formula, This refers to the output current of the photovoltaic cell. This refers to the port voltage. (Known parameters) and Calculated using the following formulas (8) and (9): (8) (9) Considering the changes in environmental conditions over time, the changes in voltage and current will also be corrected accordingly, as shown in equations (10) and (11) below: (10) (11) In the formula, , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively. , The reference irradiance and reference temperature under standard test conditions (STC) are 1000 W / m2 and 25℃, respectively.

[0031] Further consider the number of series components and parallel series number The power generation characteristic model of a marine photovoltaic array under any environment considering wave effects can be given by equation (12): (12) In the formula, For the output current of the floating photovoltaic array; The number of parallel strings; This represents the number of components connected in series. This is the short-circuit current; , All are parameters; This refers to the output voltage of the floating photovoltaic array. This is the open-circuit voltage; , These are the differentials of current and voltage, respectively.

[0032] Based on this power generation characteristic, a mathematical model can be built to simulate the generation of fault IV curves under wave conditions, which are then used as a training set for a neural network after feature extraction.

[0033] In this embodiment, in step S2, the fault IV data curve is subjected to characteristic analysis, and the open circuit voltage, short circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate are selected as fault characteristic quantities.

[0034] Specifically, considering the impact of waves on the operation of photovoltaic array power generation, different forms of faults are simulated on it. Taking a more severe local shading fault as an example, such as... Figure 3As shown, when photovoltaic (PV) modules are partially shaded, the shaded battery modules inside the PV modules cannot generate electricity, resulting in an inflection point on the IV curve and a decrease in the maximum power point. This inflection point characteristic continues to evolve with the severity of the fault over time. Simultaneously, wave effects introduce numerous inflection points with different locations and similar shapes on the IV curve, interfering with the morphological evolution of the fault inflection point. However, the inflection point phenomenon and maximum power point shift caused solely by wave effects are randomly distributed over time, without a significant and continuous deterioration trend at different times. Therefore, it is necessary to extract important characteristic variables that reflect fault information and consider the temporal evolution trend for fault diagnosis. To encompass a wide range of common offshore PV array fault types, open-circuit voltage is selected. Short-circuit current Maximum power point voltage and current Number of inflection points Inflection point voltage Inflection point current and the rate of change of the buoy's tilt angle As a fault characteristic quantity. The volatility is measured by a gyroscope device that monitors the tilt angle of the float and calculated by the monitoring host, as shown in the following formula (13): (13) In the formula, The undulation rate of the floating body's tilt angle; Let be the tilt angle of the component at the i-th sampling point; This represents the number of sampling points within that time period. This represents the average tilt angle of the photovoltaic module.

[0035] In this embodiment, in step S3, based on the floating body tilt angle volatility, the historical volatility data is clustered using the fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact.

[0036] Specifically, to fully consider the temporal evolution characteristics of faults in offshore photovoltaic arrays, this invention proposes a fault feature extraction and diagnosis strategy based on current online IV diagnostic technology. Relying on the MPPT / IV diagnostic module integrated into modern inverters, the IV curves of offshore photovoltaic arrays can be remotely and continuously acquired with a sampling time of less than 1 second, minimizing the impact on the normal power generation operation of the photovoltaic array. Therefore, remote IV diagnostics can perform continuous sampling and monitoring during operation, with a sampling interval set to 1 hour.

[0037] This invention further employs a sliding time window technique to process the feature sequence after multiple samplings on the IV curve. For a feature sequence with a window width of w and a sliding step size of s... The sequence at the nth sliding position can be represented as: (14) For multiple characteristic sequences on the IV curve of a photovoltaic array, the characteristic matrix obtained during the nth sliding period of the above window can be represented as: (15)

[0038] In this study, the window width *w* controls the amount of historical feature information retained, while the step size *s* determines the update rate of new data. Under the influence of sea wave conditions, the confusion of inflection point features makes it difficult for a fixed sliding time window to effectively capture fault characteristics. Due to significant differences in wave fluctuations across different sea areas and time periods, it is difficult to determine an optimal window parameter. Therefore, it is necessary to design an adaptive time window that can be dynamically adjusted based on the wave impact and fault evolution characteristics to determine a suitable feature sequence length. This invention employs the Fuzzy C-Means (FCM) clustering algorithm to process volatility and proposes an adaptive sliding time window adjustment strategy based on cosine similarity.

[0039] like Figure 4 As shown, to ensure that the temporal evolution information of fault characteristics is not lost during the window sliding process, and to reduce the impact on the photovoltaic array's power generation operation, this invention sets the IV curve sampling interval to 1 hour, the moving step size s to 1 hour, and the diagnostic period to an integer multiple of s. Then, the historical volatility data of the offshore photovoltaic array floating body under waves is clustered. For the volatility set of historical time periods... , Let the number of sample clusters be . The precision factor is Then the clustering objective function It can be expressed as shown in equation (16): (16) In the formula, For the sample With cluster center Membership degree; The weighted index is usually set to 2; This represents the i-th cluster center.

[0040] Initialize the membership matrix To satisfy the following conditions, as shown in equation (17): (17) Update the cluster center of each class to: (18) The updated membership matrix is ​​calculated using equation (19): (19) When satisfied When the condition is met, the algorithm stops iterating. Otherwise, continue repeating equations (18) and (19) until convergence.

[0041] In this embodiment, in step S4, the fault IV data at the current time and similar historical time are converted according to standard test conditions, and the cosine similarity of the feature vector of the converted IV curve is calculated; the degree of evolution of the fault time sequence is judged based on the cosine similarity, and the judgment result is obtained; based on the judgment result, the adaptive window width containing complete fault evolution information is determined, and the effective fault feature sequence is extracted.

[0042] Specifically, after historical volatility clustering is completed, cluster analysis can be performed on the volatility collected during the current operating period to determine the time when similar volatility influences the system. In actual operating periods, different IV curve scanning times are accompanied by wave conditions of varying strength and fault evolution. The volatility distribution at different times is random, but the fault will gradually evolve to a severe state over time. After converting multiple IV curves under the same wave influence to STC conditions, environmental interference will be eliminated. If there is no fault at this time, the multiple IV curves will be extremely similar, with a cosine similarity close to 1. However, since a fault exists at this time, and the fault gradually worsens over time, an adaptive sliding time window strategy based on the cosine similarity of the feature vectors is proposed to determine the appropriate feature sequence length to be extracted during diagnosis, such as... Figure 5 As shown.

[0043] During the continuous operation of the offshore photovoltaic array, let the eigenvector of the IV curve at time t, converted to STC conditions, be denoted as... The conversion method follows the following formulas (20) and (21): (20) (twenty one) In the formula, The current under STC conditions; The voltage under STC conditions; and These are the raw current and voltage data collected from the photovoltaic array, respectively. , These are the reference irradiance and reference temperature under standard test conditions, respectively; , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively; G is the irradiance received by the tilted surface; and T is the original photovoltaic array temperature. This is the open-circuit voltage irradiance coefficient.

[0044] After cluster analysis, compared with the current time... Let the moment when they are in the same volatility cluster be denoted as . and ,and Calculate the time respectively With time Cosine similarity of eigenvectors and time With time Cosine similarity of eigenvectors The cosine similarity between two time points is calculated by equation (22): (twenty two) In the formula, Cosine similarity; The eigenvectors of the IV curve under STC conditions; This represents the eigenvectors of IV curves at similar historical moments.

[0045] When the condition is met When this occurs, it indicates that the temporal evolution of the fault has caused significant changes in the features, and the adaptive window width can be determined as follows: h. If there are multiple times that are related to the current time... For preceding moments with similar wave conditions, repeat the calculation of formula (22) and compare until the length of the feature sequence containing complete fault feature time-series evolution information is determined. .

[0046] When the determined adaptive time window width Larger than the preset maximum window width or from the current moment Before There are no three or more times within the time step. When sampling time points with similar wave conditions are considered to be in a strongly non-stationary state during the diagnostic period, the interference information under wave motion overshadows the temporal evolution of fault characteristics. In this case, diagnosis would introduce more non-fault signals, therefore, no diagnosis is performed. Based on this adaptive sliding time window strategy, effective feature sequences containing fault information can be extracted before diagnosis and used as input to the neural network model.

[0047] In this embodiment, in step S5, a CBRM deep network model is constructed by fusing CNN, residual connection BiGRU, and MHA; the effective fault feature sequence is input into the CBRM deep network model for training and iterative optimization to obtain a trained CBRM deep network model; and the trained CBRM deep network model is used for diagnostic processing to obtain fault diagnosis results for the marine photovoltaic array.

[0048] Specifically, to capture the fault characteristics of offshore photovoltaic arrays and achieve accurate fault diagnosis, this patent employs a fused deep network model that integrates CNN, BiGRU with residual connections, and MHA, namely the CBRM model. Its processing flow is as follows: Figure 6 As shown.

[0049] First, a padding layer is introduced to zero-paste and align feature sequences of different lengths to facilitate feature learning by the neural network. Then, a two-layer convolutional CNN module is used to extract local features coupled with multivariate temporal feature variables, capturing the dependencies between variables. To fully capture the bidirectional dependencies of the operating state of the offshore photovoltaic array over time, this invention employs a BiGRU structure for global temporal feature modeling. The features extracted by the CNN are flattened and then input into the BiGRU module to fully learn the forward and backward temporal features. Let the flattened sequence output by the CNN be... The forward and backward features of BiGRU are shown in equations (23) and (24): (twenty three) (twenty four) In the formula, This indicates a forward-hidden state, reflecting the characteristics of the current moment after being influenced by history; This represents the reverse hidden state, reflecting the characteristics of the current moment after being affected by future information.

[0050] After concatenating the forward and backward temporal features, the final comprehensive temporal features are shown in formula (25): (25) To prevent the over-introduction of wave interference information during BiGRU learning, this invention introduces a parallel residual branch to preserve the original structural information extracted by the convolutional layer and enhance the model's noise resistance. The specific residual mapping is shown in formula (26): (26) In the formula, and These represent the weights and biases of the fully connected layer in the residual mapping branch, respectively.

[0051] After passing through the residual branch and BiGRU, the final fused output of the residual connection is: (27) After parallel residual branches, the initial features of the flattened layer can be dimensionally transformed by a fully connected layer, and then concatenated with the complete temporal features extracted by BiGRU in the residual connected layer. Introducing this residual structure makes the model more stable and reliable when learning temporal patterns under complex fluctuation conditions. Finally, the data is input into the multi-head attention (MHA) layer, so that the model automatically focuses on the key moments most valuable for fault diagnosis, assigning them higher weights and suppressing the interference of redundant data.

[0052] In this embodiment, considering the harsh environment in which offshore photovoltaic systems operate, it is typically difficult to obtain a large amount of fault data for model training. Therefore, this invention uses simulation data as a crucial support for network model training data. Specifically, the implementation involves: using the aforementioned mathematical model of the offshore photovoltaic array for simulation, generating IV curves of the photovoltaic array under various wave and fault conditions; establishing an adaptive sliding time window based on fuzzy C-means clustering and cosine similarity; extracting multivariate time-series fault feature matrix samples for neural network input; integrating these samples to construct a neural network training set; training the CBRM model; and finally obtaining a deep network model for offshore photovoltaic array fault diagnosis. The trained CBRM deep network model is then used for diagnostic processing to obtain the fault diagnosis results for the offshore photovoltaic array.

[0053] In one possible embodiment, a simulation verification example is provided as follows:

[0054] This invention first establishes a marine photovoltaic array power generation model in the Matlab / Simulink environment. The irradiance and environmental data used are derived from annual Bohai Sea data provided by the SolarGIS platform. The photovoltaic array used for fault simulation is configured in a 2×10 configuration, such as... Figure 7 As shown in Table 1, the parameters of the photovoltaic modules are as follows. Furthermore, the five sets of nearshore wave conditions used for testing were obtained through scaled-down model experiments, with wave heights ranging from 1m to 5m and periods ranging from 4s to 9s. To simulate complex fault conditions of offshore photovoltaic arrays, this invention combines various wave conditions with six fault states to generate IV curves, including normal operation, open circuit fault, short circuit fault, fouling fault, microcrack fault, and hot spot fault, where normal operation is considered a special fault state.

[0055] Table 1 Parameters of offshore photovoltaic modules

[0056] Based on Bohai Sea environmental data, IV curves are sampled hourly only between 9:00 AM and 4:00 PM daily. Finally, after adaptive window feature extraction, sample data for each fault type are obtained, forming a neural network sample set, which is then divided into training, validation, and test sets at a ratio of 70%, 15%, and 15%, respectively. The training set is input into the CBRM model for training and testing, yielding the fault diagnosis results for the test set, such as... Figure 8 As shown, after the CBRM model is trained, the network shows significant improvement in local feature extraction and temporal feature extraction capabilities, with an overall fault diagnosis accuracy of 98.9%. It has good recognition performance for open-circuit and short-circuit faults with transient changing characteristics, as well as contamination, hot spots, and microcracks with long-term evolution characteristics, with an overall recall and precision of over 96.4%, effectively distinguishing different fault types.

[0057] The application scenarios of this invention are as follows: This invention is applicable to floating photovoltaic power stations in various water areas, including open coastal waters, sheltered waters in harbors, inland lakes, and reservoirs. It is particularly suitable for complex operating scenarios characterized by frequent wave disturbances, severe salt spray corrosion, and the potential for fault characteristics to overlap with environmental interference. It can be directly deployed in the power station's central control system, remote operation and maintenance monitoring platform, or intelligent diagnostic terminal for routine online monitoring and fault diagnosis. It can accurately identify common fault types such as open-circuit faults, short-circuit faults, module contamination, cell microcracks, and hot spots, covering the entire lifecycle of power station construction and operation. Furthermore, it can be applied to scenarios such as operation and maintenance scheduling optimization, fault early warning and forecasting, and maintenance plan formulation for floating photovoltaic power stations, providing operation and maintenance personnel with accurate fault location and handling basis. It can also adapt to the needs of intelligent configuration in new power stations and technical upgrades and renovations of existing power stations, helping to improve the safe and stable operation level and operation and maintenance efficiency of offshore photovoltaic systems, and reducing operating costs.

[0058] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0059] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0060] Example 2

[0061] See Figure 9 Embodiment 2 of the present invention also provides a floating photovoltaic fault timing feature adaptive extraction and diagnosis device, comprising: The Fault IV data curve generation module 001 is used to obtain the time-domain tilt angle data of the floating body under the action of sea waves through the floating body scale model test; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, a marine photovoltaic array power generation characteristic model is constructed; and the fault IV data curve is generated under the set wave conditions and fault type through the marine photovoltaic array power generation characteristic model. The Fault IV Data Curve Characteristic Analysis Module 002 is used to perform characteristic analysis on the Fault IV Data Curve, and selects open-circuit voltage, short-circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate as fault characteristic quantities. The similar historical moment determination module 003 is used to determine historical moments similar to the current moment in terms of wave impact by performing cluster analysis on historical volatility data based on the floating body tilt angle volatility using a fuzzy C-means clustering algorithm. The effective fault feature sequence extraction module 004 is used to convert the fault IV data at the current time and similar historical time according to standard test conditions, calculate the cosine similarity of the feature vector of the converted IV curve; judge the degree of evolution of the fault time sequence based on the cosine similarity, and obtain the judgment result; based on the judgment result, determine the adaptive window width containing complete fault evolution information, and extract the effective fault feature sequence. The CBRM deep network model construction and processing module 005 is used to construct a CBRM deep network model by fusing CNN, residual connection BiGRU and MHA; input the effective fault feature sequence into the CBRM deep network model for training and iterative optimization to obtain a trained CBRM deep network model; and perform diagnostic processing through the trained CBRM deep network model to obtain fault diagnosis results for marine photovoltaic arrays.

[0062] In this embodiment, the expression for the offshore photovoltaic array power generation characteristic model in the fault IV data curve generation module 001 is as follows: ; In the formula, For the output current of the floating photovoltaic array; The number of parallel strings; This represents the number of components connected in series. This is the short-circuit current; , All are parameters; This refers to the output voltage of the floating photovoltaic array. This is the open-circuit voltage; , These are the differentials of current and voltage, respectively.

[0063] In this embodiment, in the fault IV data curve characteristic analysis module 002, the float tilt angle fluctuation rate is tested by a gyroscope device that monitors the float tilt angle and calculated by the monitoring host; the expression for the float tilt angle fluctuation rate is: ; In the formula, The undulation rate of the floating body's tilt angle; Let be the tilt angle of the component at the i-th sampling point; This represents the number of sampling points within that time period. This represents the average tilt angle of the photovoltaic module.

[0064] In this embodiment, in the effective fault feature sequence extraction module 004, during the process of converting the fault IV data of the current time and similar historical times according to standard test conditions, the conversion formula is as follows: ; ; In the formula, The current under STC conditions; The voltage under STC conditions; and These are the raw current and voltage data collected from the photovoltaic array, respectively. , These are the reference irradiance and reference temperature under standard test conditions, respectively; , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively; G is the irradiance received by the tilted surface; and T is the original photovoltaic array temperature. This is the open-circuit voltage irradiance coefficient; The formula for calculating the cosine similarity of the eigenvectors of the IV curves after conversion is as follows: ; In the formula, Cosine similarity; The eigenvectors of the IV curve under STC conditions; This represents the eigenvectors of IV curves at similar historical moments.

[0065] In this embodiment, in the effective fault feature sequence extraction module 004, during the process of determining the adaptive window width containing complete fault evolution information based on the judgment result, if the adaptive window width is greater than the set maximum window width, or the number of similar historical sampling points is less than 3, it is determined to be a strongly non-stationary state, and diagnosis is not performed temporarily.

[0066] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0067] Example 3

[0068] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for an adaptive extraction and diagnosis method of timing features of floating photovoltaic faults. The program code includes instructions for executing the adaptive extraction and diagnosis method of timing features of floating photovoltaic faults according to Embodiment 1 or any possible implementation thereof.

[0069] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0070] Example 4

[0071] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor; The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute the adaptive extraction and diagnosis method for timing features of floating photovoltaic faults according to Embodiment 1 or any possible implementation thereof.

[0072] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0073] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0074] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0075] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. An adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults, characterized in that, include: Through scaled-down model tests of the floating body, time-domain tilt angle data of the floating body under the action of sea waves were obtained; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, a marine photovoltaic array power generation characteristic model was constructed; through the marine photovoltaic array power generation characteristic model, fault IV data curves under set wave conditions and fault types were generated. The fault IV data curve was subjected to characteristic analysis, and the open circuit voltage, short circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate were selected as fault characteristic quantities. Based on the floating body tilt angle volatility, the historical volatility data is clustered using the fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact. The fault IV data at the current time and similar historical time are converted according to standard test conditions, and the cosine similarity of the feature vector of the converted IV curve is calculated. The degree of evolution of the fault time sequence is judged based on the cosine similarity, and the judgment result is obtained. Based on the judgment result, the adaptive window width containing complete fault evolution information is determined, and the effective fault feature sequence is extracted. A deep network model for CBRM is constructed by fusing CNN, residual connection BiGRU, and MHA; the effective fault feature sequence is input into the deep network model for training and iterative optimization to obtain a trained deep network model; the trained deep network model is then used for diagnostic processing to obtain fault diagnosis results for marine photovoltaic arrays.

2. The adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults according to claim 1, characterized in that, The expression for the power generation characteristic model of the offshore photovoltaic array is: ； In the formula, For the output current of the floating photovoltaic array; The number of parallel strings; This represents the number of components connected in series. This is the short-circuit current; , All are parameters; This refers to the output voltage of the floating photovoltaic array. This is the open-circuit voltage; , These are the differentials of current and voltage, respectively.

3. The adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults according to claim 2, characterized in that, The float tilt angle fluctuation rate is measured by a gyroscope device that monitors the float tilt angle and calculated by the monitoring host; the expression for the float tilt angle fluctuation rate is: ； In the formula, The undulation rate of the floating body's tilt angle; Let be the tilt angle of the component at the i-th sampling point; This represents the number of sampling points within that time period. This represents the average tilt angle of the photovoltaic module.

4. The adaptive extraction and diagnosis method for the timing features of floating photovoltaic faults according to claim 3, characterized in that, In the process of converting the current fault IV data with similar historical times according to standard test conditions, the conversion formula is as follows: ； ； In the formula, The current under STC conditions; The voltage under STC conditions; and These are the raw current and voltage data collected from the photovoltaic array, respectively. , These are the reference irradiance and reference temperature under standard test conditions, respectively. , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively; G is the irradiance received by the tilted surface; and T is the original photovoltaic array temperature. This is the open-circuit voltage irradiance coefficient; The formula for calculating the cosine similarity of the eigenvectors of the IV curves after conversion is as follows: ； In the formula, Cosine similarity; The eigenvectors of the IV curve under STC conditions; This represents the eigenvectors of IV curves at similar historical moments.

5. The adaptive extraction and diagnosis method for timing features of floating photovoltaic faults according to claim 4, characterized in that, In the process of determining the adaptive window width containing complete fault evolution information based on the judgment result, if the adaptive window width is greater than the set maximum window width, or if there are fewer than 3 similar historical sampling points, it is determined to be a strongly non-stationary state and diagnosis will not be performed for the time being.

6. A floating photovoltaic fault timing feature adaptive extraction and diagnosis device, employing the floating photovoltaic fault timing feature adaptive extraction and diagnosis method according to any one of claims 1-5, characterized in that, include: The Fault IV data curve generation module is used to obtain the time-domain tilt angle data of the floating body under the action of sea waves through the floating body scale model test; based on the time-domain tilt angle data of the floating body, combined with the photovoltaic cell engineering model, a marine photovoltaic array power generation characteristic model is constructed; and the Fault IV data curve is generated under the set wave conditions and fault type through the marine photovoltaic array power generation characteristic model. The Fault IV data curve characteristic analysis module is used to perform characteristic analysis on the Fault IV data curve, and selects open circuit voltage, short circuit current, maximum power point voltage, maximum power point current, number of inflection points, inflection point voltage, inflection point current and float tilt angle fluctuation rate as fault characteristic quantities. The similar historical moment determination module is used to perform cluster analysis on historical volatility data based on the floating body tilt angle volatility using a fuzzy C-means clustering algorithm to determine historical moments similar to the current moment's wave impact. The effective fault feature sequence extraction module is used to convert fault IV data at the current time and similar historical time according to standard test conditions, calculate the cosine similarity of the feature vector of the converted IV curve; judge the degree of evolution of the fault time sequence based on the cosine similarity, and obtain the judgment result; based on the judgment result, determine the adaptive window width containing complete fault evolution information, and extract the effective fault feature sequence. The CBRM deep network model construction and processing module is used to construct a CBRM deep network model by fusing CNN, residual connection BiGRU and MHA; input the effective fault feature sequence into the CBRM deep network model for training and iterative optimization to obtain a trained CBRM deep network model; and perform diagnostic processing through the trained CBRM deep network model to obtain fault diagnosis results for marine photovoltaic arrays.

7. The floating photovoltaic fault timing feature adaptive extraction and diagnostic device according to claim 6, characterized in that, In the fault IV data curve generation module, the expression for the offshore photovoltaic array power generation characteristic model is: ； In the formula, For the output current of the floating photovoltaic array; The number of parallel strings; This represents the number of components connected in series. This is the short-circuit current; , All are parameters; This refers to the output voltage of the floating photovoltaic array. This is the open-circuit voltage; , These are the differentials of current and voltage, respectively.

8. The floating photovoltaic fault timing feature adaptive extraction and diagnostic device according to claim 7, characterized in that, In the fault IV data curve characteristic analysis module, the float tilt angle fluctuation rate is measured by a gyroscope device that monitors the float tilt angle and calculated by the monitoring host; the expression for the float tilt angle fluctuation rate is: ； In the formula, The undulation rate of the floating body's tilt angle; Let be the tilt angle of the component at the i-th sampling point; This represents the number of sampling points within that time period. This represents the average tilt angle of the photovoltaic module.

9. The floating photovoltaic fault timing feature adaptive extraction and diagnostic device according to claim 8, characterized in that, In the effective fault feature sequence extraction module, during the process of converting the fault IV data at the current time and similar historical times according to standard test conditions, the conversion formula is as follows: ； ； In the formula, The current under STC conditions; The voltage under STC conditions; and These are the raw current and voltage data collected from the photovoltaic array, respectively. , These are the reference irradiance and reference temperature under standard test conditions, respectively. , These are the temperature coefficients of short-circuit current and open-circuit voltage, respectively; G is the irradiance received by the tilted surface; and T is the original photovoltaic array temperature. This is the open-circuit voltage irradiance coefficient; The formula for calculating the cosine similarity of the eigenvectors of the IV curves after conversion is as follows: ； In the formula, Cosine similarity; The eigenvectors of the IV curve under STC conditions; This represents the eigenvectors of IV curves at similar historical moments.

10. The floating photovoltaic fault timing feature adaptive extraction and diagnostic device according to claim 9, characterized in that, In the effective fault feature sequence extraction module, during the process of determining the adaptive window width containing complete fault evolution information based on the judgment result, if the adaptive window width is greater than the set maximum window width, or if there are fewer than 3 similar historical sampling points, it is determined to be a strongly non-stationary state and diagnosis is temporarily not performed.

Images