Photovoltaic array fault diagnosis method based on benchmark multi-dimensional deviation calculation

Abstract

The invention provides a photovoltaic array fault diagnosis method based on benchmark multi-dimensional deviation calculation, and the method aims at the defects that traditional diagnosis depends on a fixed reference and a fault knowledge base is static, and comprises the steps: 1, constructing a dynamic benchmark array, combining equipment installation information and real-time operation data, screening or constructing a combined reference according to a performance ratio mean value and variance, and constructing a dynamic benchmark array; the reference interference of assembly aging and environment change is eliminated; 2, quantitatively calculating deviations of key physical quantities such as power, voltage and current, and standardizing the deviations to form a deviation vector with high distinction degree; 3, establishing a self-updating fault knowledge base, generating a fault rule through historical sample training, mapping a fault mode, and supporting novel fault feature supplementation; and 4, introducing characteristic quantity calculation and distance threshold determination, eliminating characteristic correlation interference in combination with a covariance matrix, and additionally adding an unknown fault processing mechanism and an operation and maintenance linkage module. The method is suitable for photovoltaic power stations of various scales, and is especially suitable for photovoltaic array operation and maintenance in complex environments such as mountainous areas and deserts.

Description

Technical Field

[0001] This invention relates to the field of new energy operation and maintenance technology, specifically to a photovoltaic array fault diagnosis method based on benchmark multidimensional deviation calculation. Background Technology

[0002] In the operation and maintenance of photovoltaic power plants, fault diagnosis of photovoltaic arrays is a core link in ensuring the efficient operation of the power plant. Existing technologies for photovoltaic fault diagnosis suffer from two major problems: First, traditional diagnostic methods often rely on fixed benchmarks or empirical thresholds, making it difficult to adapt to performance drift caused by component aging and dynamic environmental changes in photovoltaic arrays. This results in a lack of reference benchmarks and easily leads to diagnostic biases. Second, existing fault knowledge bases are mostly statically preset, unable to autonomously learn new fault types or update existing fault characteristics. This makes it difficult to cope with new faults or the evolution of fault modes that occur during long-term operation, leading to incomplete fault coverage and a decrease in diagnostic accuracy over time.

[0003] To address the aforementioned issues, this invention proposes a photovoltaic array fault diagnosis method based on dynamic benchmark multidimensional deviation calculation. This method provides a reference for photovoltaic array fault diagnosis by constructing dynamic benchmarks, enhances the distinguishability of fault features through multidimensional deviation quantification, and establishes a fault knowledge base. Finally, fault diagnosis is achieved by comparing the fault with different faults in the fault knowledge base, significantly improving the accuracy and robustness of photovoltaic array fault diagnosis. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a photovoltaic array fault diagnosis method based on benchmark multidimensional deviation calculation, which achieves: benchmark adaptation: dynamically tracking array performance changes and eliminating the influence of aging and environmental interference on the benchmark; accurate fault identification: improving the distinguishability of fault features through multidimensional deviation quantification and reducing the false and false judgment rates.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A photovoltaic array fault diagnosis method based on benchmark multidimensional deviation calculation includes the following steps: S1. Dynamic benchmark array determination: Collect equipment installation information and operation data of multiple photovoltaic arrays in the photovoltaic power station, and determine one or more benchmark arrays based on the performance stability and theoretical power deviation of each photovoltaic array; S2. Multidimensional deviation quantification calculation: Compare the operating data of the array to be diagnosed and the benchmark array at the same time point, quantify the deviation value of key physical quantities, and standardize the deviation value to form a standardized deviation vector. S3. Fault Knowledge Base Training: A fault knowledge base is established using historical deviation data and statistical analysis combined with fault mode mapping. S4. Real-time fault diagnosis: Real-time input of the standardized deviation vector of the array to be diagnosed. Fault diagnosis is achieved by comparing the faults with the fault category features in the fault knowledge base.

[0006] The above-mentioned S1 specifically includes the following steps: S11. Data Acquisition: Obtain equipment installation information and real-time operation data of the photovoltaic array. Equipment installation information includes rated power of the components, number of strings, number of parallel connections, and installation years. Real-time operation data includes output power, ambient temperature, irradiance, operating voltage, and operating current. S12. Theoretical Power Calculation: Based on the characteristics of photovoltaic modules, calculate the rated power of the photovoltaic array under standard conditions, and make corrections based on the actual operating environment to obtain the theoretical power. The formula for calculating the theoretical power is: ; in The power temperature coefficient of the component. This refers to the actual ambient temperature. For standard test temperature, This represents the actual irradiance. Standard test irradiance; S13. Benchmark Array Selection: Calculate the mean and variance of the performance ratio of each photovoltaic array within a preset statistical period, and select the photovoltaic array with the mean performance ratio closest to 1 and the smallest variance as the benchmark array; if the stability of a single array is insufficient, select multiple photovoltaic arrays with the best performance and construct a joint benchmark array based on the mean of their operating data.

[0007] In S2 above, the deviation values ​​of key physical quantities include: Power deviation: ; Voltage deviation: ; Current deviation: ; Standardized power deviation: .

[0008] The aforementioned power deviation Voltage deviation Current deviation and standardized power deviation Constructing the standardized deviation vector .

[0009] The above-mentioned S3 specifically includes the following steps: S31. Sample Construction: This involves constructing multiple standardized deviation vectors from the collected samples. The training sample set is formed by the corresponding fault labels. ,in The fault labels include "component mismatch", "obstruction", "diode failure" and "poor cable contact". S32. Fault rule generation: For the training sample set... Statistical analysis is performed to extract the deviation feature rules corresponding to various types of faults. The deviation feature rules are characterized by the numerical range of various types of deviations. S33. Fault Mode Mapping: Extracted deviation feature rules are mapped one-to-one with actual fault modes to establish an initial fault knowledge base. This initial fault knowledge base stores the feature mean values ​​of various fault types. covariance matrix And the deviation feature rule description, where k is the fault category number.

[0010] The above-mentioned S4 specifically includes the following steps: S41. Feature Calculation: For the standardized deviation vector acquired in real time... Calculate its feature quantity relative to each fault category k in the fault knowledge base: ; S42. Fault determination: Preset distance threshold If a certain fault category exists If the array is classified as faulty, then it is determined to belong to the k-th type of fault; if all types are greater than the threshold... If the error is not found, it is determined to be an "unknown fault" and the corresponding deviation vector is sent to the unknown fault candidate pool.

[0011] In the above-mentioned S11, data acquisition uses an edge computing gateway, the sampling frequency is set to 1 min / time, the data transmission delay is ≤500ms, and outliers are removed from the collected real-time running data. The removal rule is: when a single set of data exceeds the mean of the same batch of data ±3 times the standard deviation, it is judged as an outlier and is filled by linear interpolation.

[0012] The average operating data of the joint benchmark array in S13 above is calculated as follows: ; ; ; in, n The number of arrays participating in the construction of the joint benchmark, , , These are the output power, operating voltage, and operating current of the i-th array, respectively. , The variance of the performance ratio of the i-th array, weights This is used to highlight the proportion of array data with more stable performance.

[0013] The aforementioned S3 also includes: S34. Knowledge base update: Collect new fault samples every 3 months, and update the feature mean, covariance matrix and deviation feature rules in the fault knowledge base through incremental learning. Add fault type annotations and include them in the knowledge base. Incremental learning employs a few-shot learning model. When the number of newly added fault samples is ≥50, an update is initiated. During the update process, the original fault feature data is retained, and only the fault category parameters corresponding to the newly added samples are adjusted to ensure the compatibility of the knowledge base.

[0014] The aforementioned unknown fault candidate pool is equipped with a periodic review mechanism. Every month, the deviation vectors in the pool are clustered and analyzed. If the number of vectors in a certain cluster is ≥30 and the corresponding fault type is confirmed by manual investigation, then the vectors in that cluster are used as new fault samples, and the knowledge base update process in S34 is executed.

[0015] The method also includes a fault early warning step S5: real-time monitoring of the standardized deviation vector change trend of the array to be diagnosed, when within 5 consecutive sampling periods... The decrease is ≥0.05 or When the absolute value increases by ≥5V, a fault warning is triggered, and the warning level and pre-processing suggestions are output; the warning level includes mild / moderate / severe.

[0016] The criteria for determining the warning level in S5 mentioned above are as follows: Mild warning: The deviation change range is 1-1.5 times the warning threshold for 5 consecutive periods, and the current... ; Moderate warning: The deviation change range is 1.5-2 times the warning threshold for 5 consecutive periods, and the current... ; Severe warning: The deviation change is ≥ 2 times the warning threshold within 5 consecutive periods, and the current... .

[0017] The aforementioned dynamic benchmark array is re-screened every 7 days. If the performance ratio of the original benchmark array is still within the range of 0.98-1.02 and the variance is ≤0.015 during the screening, it is retained as a benchmark array. If the performance of the original benchmark array deteriorates, the screening process of S12-S13 is repeated to ensure the effectiveness of the benchmark array.

[0018] The calculation of the voltage and current deviations mentioned above needs to be synchronized with the working periods of the array to be diagnosed and the benchmark array. Only data from the same time point with irradiance ≥300W / m² and ambient temperature between -10℃ and 45℃ are selected for comparison to exclude the interference of low irradiance and extreme temperature on the deviation calculation.

[0019] The aforementioned S4 also includes: S43. Output Results: Output the fault diagnosis results, including fault category and fault confidence level. We recommend checking the affected areas and simultaneously updating the diagnostic results to the photovoltaic power station operation and maintenance management platform. The fault confidence level is divided into three levels: high confidence level: confidence level ≥ 0.8, medium confidence level: 0.5 ≤ confidence level < 0.8, and low confidence level: confidence level < 0.5. When the confidence level is low, the secondary diagnosis is automatically started, and three sets of real-time data are collected again to calculate the standardized deviation vector. The mean of the three feature values ​​is taken as the final judgment basis.

[0020] The photovoltaic array fault diagnosis method based on benchmark multidimensional deviation calculation mentioned in this invention has the following beneficial effects: 1. High diagnostic accuracy: Dynamic benchmarks eliminate reference interference, multi-dimensional deviations improve feature discrimination, Mahalanobis distance calculation eliminates correlation effects, fault diagnosis accuracy ≥94%, misjudgment rate ≤5%, and missed judgment rate ≤3%.

[0021] 2. High adaptability: Supports switching between single array and joint benchmark, adapting to component aging and environmental change scenarios; the knowledge base is self-updated to support new fault identification, and the diagnostic accuracy rate remains ≥92% after 5 years of operation.

[0022] 3. High operation and maintenance efficiency: The new fault early warning mechanism can identify fault risks on average 2-3 days in advance; the diagnosis results are synchronized with the operation and maintenance platform, the average fault handling time is shortened to 8 minutes / fault, and the operation and maintenance cost is reduced by 40%-50%.

[0023] 4. Wide range of applications: It is suitable for photovoltaic power stations in different scenarios such as mountainous areas, deserts, and rooftops, and supports the deployment of power stations of different scales from 10 to 1000MW, with strong compatibility. Attached Figure Description

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 : A schematic diagram of the overall process of the photovoltaic array fault diagnosis method of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0026] Implementation 1: A photovoltaic array fault diagnosis method based on benchmark multidimensional deviation calculation includes the following steps: S1. Dynamic benchmark array determination: Collect equipment installation information and operation data of multiple photovoltaic arrays in the photovoltaic power station, and determine one or more benchmark arrays based on the performance stability and theoretical power deviation of each photovoltaic array; S2. Multidimensional deviation quantification calculation: Compare the operating data of the array to be diagnosed and the benchmark array at the same time point, quantify the deviation value of key physical quantities, and standardize the deviation value to form a standardized deviation vector. S3. Fault Knowledge Base Training: A fault knowledge base is established using historical deviation data and statistical analysis combined with fault mode mapping. S4. Real-time fault diagnosis: Real-time input of the standardized deviation vector of the array to be diagnosed. Fault diagnosis is achieved by comparing the faults with the fault category features in the fault knowledge base.

[0027] The above-mentioned S1 specifically includes the following steps: S11. Data Acquisition: Obtain equipment installation information and real-time operation data of the photovoltaic array. Equipment installation information includes rated power of the components, number of strings, number of parallel connections, and installation years. Real-time operation data includes output power, ambient temperature, irradiance, operating voltage, and operating current. S12. Theoretical Power Calculation: Based on the characteristics of photovoltaic modules, calculate the rated power of the photovoltaic array under standard conditions, and make corrections based on the actual operating environment to obtain the theoretical power. The formula for calculating the theoretical power is: ; in The power temperature coefficient of the component. This refers to the actual ambient temperature. The standard test temperature is 25℃. This represents the actual irradiance. Standard test irradiance: 1000 W / m²; S13. Benchmark Array Selection: Calculate the mean and variance of the performance ratio of each photovoltaic array within a preset statistical period, and select the photovoltaic array with the mean performance ratio closest to 1 and the smallest variance as the benchmark array; if the stability of a single array is insufficient, select multiple photovoltaic arrays with the best performance and construct a joint benchmark array based on the mean of their operating data.

[0028] In S2 above, the deviation values ​​of key physical quantities include: power deviation: ; Voltage deviation: ; Current deviation: ; Standardized power deviation: .

[0029] The aforementioned power deviation Voltage deviation Current deviation and standardized power deviation Constructing the standardized deviation vector .

[0030] The above-mentioned S3 specifically includes the following steps: S31. Sample Construction: This involves constructing multiple standardized deviation vectors from the collected samples. The training sample set is formed by the corresponding fault labels. ,in The fault labels include "component mismatch", "obstruction", "diode failure" and "poor cable contact". S32. Fault rule generation: For the training sample set... Statistical analysis is performed to extract the deviation feature rules corresponding to various types of faults. The deviation feature rules are characterized by the numerical range of various types of deviations. S33. Fault Mode Mapping: Extracted deviation feature rules are mapped one-to-one with actual fault modes to establish an initial fault knowledge base. This initial fault knowledge base stores the feature mean values ​​of various fault types. covariance matrix And the deviation feature rule description, where k is the fault category number.

[0031] The above-mentioned S4 specifically includes the following steps: S41. Feature Calculation: For the standardized deviation vector acquired in real time... Calculate its feature quantity relative to each fault category k in the fault knowledge base: ; S42. Fault determination: Preset distance threshold If a certain fault category exists If the array is classified as faulty, then it is determined to belong to the k-th type of fault; if all types are greater than the threshold... If the error is not found, it is determined to be an "unknown fault" and the corresponding deviation vector is sent to the unknown fault candidate pool.

[0032] In the above-mentioned S11, data acquisition uses an edge computing gateway, the sampling frequency is set to 1 min / time, the data transmission delay is ≤500ms, and outliers are removed from the collected real-time running data. The removal rule is: when a single set of data exceeds the mean of the same batch of data ±3 times the standard deviation, it is judged as an outlier and is filled by linear interpolation.

[0033] The average operating data of the joint benchmark array in S13 above is calculated as follows: ; ; ; in, n The number of arrays participating in the construction of the joint benchmark, , , These are the output power, operating voltage, and operating current of the i-th array, respectively. , The variance of the performance ratio of the i-th array, weights This is used to highlight the proportion of array data with more stable performance.

[0034] The aforementioned S3 also includes: S34. Knowledge base update: Collect new fault samples every 3 months, and update the feature mean, covariance matrix and deviation feature rules in the fault knowledge base through incremental learning. Add fault type annotations and include them in the knowledge base. Incremental learning employs a few-shot learning model, such as Few-Shot Learning. When the number of newly added fault samples is ≥50, an update is initiated. During the update process, the original fault feature data is retained, and only the fault category parameters corresponding to the newly added samples are adjusted to ensure the compatibility of the knowledge base.

[0035] The aforementioned unknown fault candidate pool is equipped with a periodic review mechanism. Every month, the deviation vectors in the pool are clustered and analyzed using the K-Means algorithm. If the number of vectors in a certain cluster is ≥30 and the corresponding fault type is confirmed by manual investigation, then the vectors in that cluster are used as new fault samples, and the knowledge base update process in S34 is executed.

[0036] The method also includes a fault early warning step S5: real-time monitoring of the standardized deviation vector change trend of the array to be diagnosed, when within 5 consecutive sampling periods... The decrease is ≥0.05 or When the absolute value increases by ≥5V, a fault warning is triggered, and the warning level and pre-processing suggestions are output; the warning level includes mild / moderate / severe.

[0037] The criteria for determining the warning level in S5 mentioned above are as follows: Mild warning: The deviation change range is 1-1.5 times the warning threshold for 5 consecutive periods, and the current... ; Moderate warning: The deviation change range is 1.5-2 times the warning threshold for 5 consecutive periods, and the current... ; Severe warning: The deviation change is ≥ 2 times the warning threshold within 5 consecutive periods, and the current... .

[0038] The aforementioned dynamic benchmark array is re-screened every 7 days. If the performance ratio of the original benchmark array is still within the range of 0.98-1.02 and the variance is ≤0.015 during the screening, it is retained as a benchmark array. If the performance of the original benchmark array deteriorates, the screening process of S12-S13 is repeated to ensure the effectiveness of the benchmark array.

[0039] The calculation of the voltage and current deviations mentioned above needs to be synchronized with the working periods of the array to be diagnosed and the benchmark array. Only data from the same time point with irradiance ≥300W / m² and ambient temperature between -10℃ and 45℃ are selected for comparison to exclude the interference of low irradiance and extreme temperature on the deviation calculation.

[0040] The aforementioned S4 also includes: S43. Output Results: Output the fault diagnosis results, including fault category and fault confidence level. We recommend checking the affected areas and simultaneously updating the diagnostic results to the photovoltaic power station operation and maintenance management platform. The fault confidence level is divided into three levels: high confidence level: confidence level ≥ 0.8, medium confidence level: 0.5 ≤ confidence level < 0.8, and low confidence level: confidence level < 0.5. When the confidence level is low, the secondary diagnosis is automatically started, and three sets of real-time data are collected again to calculate the standardized deviation vector. The mean of the three feature values ​​is taken as the final judgment basis.

[0041] The technical solution of this invention is achieved through the following four core steps: Dynamic benchmark array determines S1: The core is to build a benchmark that "adapts to changes over time and with the environment," avoiding the drawbacks of a fixed benchmark. Data Acquisition S11: Data acquisition is adopted with a sampling frequency of 1 minute / time to synchronously acquire equipment installation information, such as component model XX-270, rated power 270W, number of series 18, number of parallel 20, installation period 2 years, and real-time operating data, such as output power 65-75kW, ambient temperature -5℃-40℃, and irradiance 200-1000W / m². Theoretical power calculation S12: Taking a certain array as an example, when the ambient temperature is 30℃ and the irradiance is 800W / m², substitute the values ​​into the formula to calculate the theoretical power: P theory = 270W×18×20×[1+(-0.4% / ℃)×(30-25℃)]×(800 / 1000) = 76.61kW, which reflects the array's "ideal power to be achieved"; Benchmark Screening S13: Statistically calculate the PR values ​​of each array over 30 days. For example, the average PR value of array 3 is 1.01 and the variance is 0.01, which is closest to 1 and stable, so it is selected as a single benchmark. If the PR variance of array 3 suddenly increases to 0.06 in a certain week, then combine array 1 (average PR value 0.98, variance 0.02) and array 4 (average PR value 0.97, variance 0.02) and calculate the average power of the three, such as 75kW + 72kW + 71kW = 72.67kW, and use it as a joint benchmark.

[0042] Multidimensional deviation quantization calculation S2: By using "deviation calculation + standardization" to highlight fault characteristics and eliminate environmental interference: Taking the array to be diagnosed 10 as an example, at the same time point, the power of the benchmark array 3 is 75kW, the voltage is 520V, and the current is 14.42A, while the power of the array to be diagnosed is 65kW, the voltage is 500V, and the current is 13A. Deviation calculation: ΔP=65-75=-10kW, ΔU=500-520=-20V, ΔI=13-14.42=-1.42A; Standardization: ΔP norm =-10 / [75×(800 / 1000)]=-0.167, eliminating the influence of irradiance on power deviation; The vector formed is Δ(t) = [-0.167, -20V, -1.42A]. This vector directly reflects the difference between the array to be diagnosed and the "ideal state".

[0043] Fault Knowledge Base Training S3: Build a knowledge base based on "data-driven + rule mapping" to provide a basis for diagnosis: Sample construction: Collect 1000 sets of historical data, including 300 sets of "occlusion" and 200 sets of "component mismatch", to form sample pairs, i.e., Δ(t), fault labels; Rule generation: ΔP of "occlusion fault" is discovered through statistical analysis. norm The values ​​are mostly between -0.3 and -0.1, and ΔU is mostly between -25V and -10V, forming quantifiable fault rules; Pattern mapping: Binds rules to fault types and stores the μ of the "occlusion" class. k =[-0.2, -18V, -1.2A]、Σ k (3×3 covariance matrix, such as Cov(ΔP)) norm (ΔU)=2.5), complete the initial knowledge base construction.

[0044] Real-time fault diagnosis S4: Accurate diagnosis is achieved based on feature quantity calculation, with the advantage of eliminating interference from correlations between features: the array to be diagnosed Δ(t) = [-0.167, -20V, -1.42A] and the "occlusion" class μ k =[-0.2, -18V, -1.2A] Calculate the characteristic quantities, and obtain D. k ≈1.2; Fault diagnosis: Preset D th =2.0, since 1.2 < 2.0, it is determined to be an "occlusion" fault; if all D of a certain array k If the value is greater than 2.0, it is determined to be an "unknown fault".

[0045] Example 2: Dynamic benchmark array determines S1: Data Acquisition S11: From January 1st to January 30th, 2024, collect equipment information (e.g., array 3: installation period 2 years, component rated power 270W, 18 series × 20 parallel) and real-time operating data (e.g., at 12:00 on January 15th, array 3 output power 75kW, ambient temperature 30℃, irradiance 800W / m²). Theoretical power calculation S12: α of array 3 t =-0.4% / ℃, Theoretical power at 12:00 on January 15th: P theory =270×18×20×[1+(-0.4%)×(30-25)]×(800 / 1000)=76.61kW; Benchmark Screening S13: Calculate the mean and variance of PR for each array from January 1st to January 30th: Array 3: PR mean = 75kW / 76.61kW ≈ 0.98, variance = 0.01; Array 5: PR mean = 73kW / 75.2kW ≈ 0.97, variance = 0.02; Array 8: PR mean = 77kW / 76.8kW ≈ 1.00, variance = 0.015; Array 3 (with the PR mean closest to 1 and the smallest variance) was selected as a single benchmark array.

[0046] Multidimensional deviation quantization calculation S2: At 10:00 AM on February 1, 2024, the array to be diagnosed was array 10. Real-time data: Array 10 output power 65kW, operating voltage 500V, operating current 13A, component surface temperature 32℃; Benchmark data: Array 3 output power 75kW, operating voltage 520V, operating current 14.42A, component surface temperature 30℃; Deviation calculation: ΔP = 65 - 75 = -10kW, ΔU = 500 - 520 = -20V, ΔI = 13 - 14.42 = -1.42A, ΔT s ᵤᵣ=32-30=2℃; Standardization: ΔP norm =-10 / [75×(800 / 1000)]=-0.167; Deviation vector: Δ(t) = [-0.167, -20V, -1.42A, 2℃].

[0047] Fault Knowledge Base Training S3 Sample construction: 1000 sets of historical data were collected from January 2023 to December 2023, including 300 sets of "shading", 200 sets of "component mismatch", 250 sets of "diode failure" and 250 sets of "poor cable contact". Fault rule generation: Rules are extracted through statistical analysis. Occlusion: ΔP norm ∈[-0.3, -0.1], ΔU∈[-25V, -10V], ΔI∈[-2A, -0.5A], ΔT sur ∈[1℃, 5℃]; Component mismatch: ΔP norm ∈[-0.1, -0.05], ΔU∈[-10V, 5V], ΔI∈[-0.5A, 0A], ΔT sur ∈[-2℃, 2℃]; Fault mode mapping: storing μ of the "occlusion" class k =[-0.2, -18V, -1.2A, 3℃], Σ k It is a 4×4 covariance matrix, such as Cov(ΔP) norm ,ΔT sur =0.2, completing the initial knowledge base V1.0.

[0048] Real-time fault diagnosis S4 Distance calculation: Array to be diagnosed Δ(t) = [-0.167, -20V, -1.42A, 2℃] and "Obstruction" type μ k =[-0.2, -18V, -1.2A, 3℃], Calculate the characteristic quantity D k ≈1.2; Fault diagnosis: D th=2.0, 1.2<2.0, the array 10 was determined to be "occlusion" fault. On-site investigation found that three components of array 10 were blocked by trees, which was consistent with the diagnosis.

Claims

1. A photovoltaic array fault diagnostic method based on benchmarking multi-dimensional deviation calculation, characterized in that, It comprises the following steps: S1, dynamic benchmark array determination: collecting equipment installation information and operation data of a plurality of photovoltaic arrays in a photovoltaic power station, and determining one or more benchmark arrays according to the performance stability and theoretical power deviation of each photovoltaic array; S2, multi-dimensional deviation quantitative calculation: comparing the operation data of the array to be diagnosed with the benchmark array at the same time point, quantitatively calculating the deviation value of the key physical quantity, and standardizing the deviation value to form a standardized deviation vector; S3, fault knowledge base training: using historical deviation data, establishing a fault knowledge base through statistical analysis combined with fault mode mapping; S4, Real-time fault diagnosis: input the standardized deviation vector of the array to be diagnosed in real time , and realize fault diagnosis by comparing with the fault category characteristics in the fault knowledge base.

2. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 1, characterized in that, The S1 specifically comprises the following steps: S11, data acquisition: obtaining equipment installation information and real-time operation data of the photovoltaic array, the equipment installation information including component rated power, string number, parallel number, installation age, and the real-time operation data including output power, environmental temperature, irradiance, working voltage and working current; S12, theoretical power calculation: calculating the rated power of the photovoltaic array under standard conditions according to the characteristics of the photovoltaic component, and correcting it combined with the actual operating environment to obtain the theoretical power, the calculation formula of the theoretical power being: ； wherein is the component power temperature coefficient, is the actual ambient temperature, is the standard test temperature, is the actual irradiance, is the standard test irradiance; S13, benchmark array screening: calculating the performance ratio mean and variance of each photovoltaic array in a preset statistical period, and screening the photovoltaic array with the performance ratio mean closest to 1 and the smallest variance as the benchmark array; if the stability of a single array is insufficient, a plurality of photovoltaic arrays with the best performance are selected to construct a joint benchmark array with the mean of their operation data.

3. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 1, characterized in that, In the S2, the deviation value of the key physical quantity includes: Power deviation: ； Voltage deviation: ； Current deviation: ； Standardized power deviation: 。 4. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 3, characterized in that, the power deviation , the voltage deviation , the current deviation and the normalized power deviation constitute a normalized deviation vector .

5. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 1, characterized in that, The S3 specifically comprises the following steps: S31, sample construction: the plurality of standardized deviation vectors collected and the corresponding fault labels constitute a training sample set wherein is a fault label, and the fault label includes "component mismatch", "occlusion", "diode failure", and "poor cable contact". S32, fault rule generation: statistical analysis is performed on the training sample set to extract the deviation feature rules corresponding to various faults, and the deviation feature rules are characterized by the numerical ranges of various deviations. S33, fault mode mapping: one-to-one correspondence between the extracted deviation feature rules and the actual fault mode, establishing an initial fault knowledge base, the initial fault knowledge base stores the feature mean value of each type of fault , covariance matrix and deviation feature rule description, where k is the fault category number.

6. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 1, characterized in that, The S4 specifically comprises the following steps: S41, Feature quantity calculation: for the standardized deviation vector acquired in real time , calculate its feature quantity with each fault category k in the fault knowledge base: ； S42, fault determination: preset distance threshold If there is a certain fault category , it is determined that the array belongs to the kth fault category; if all categories are greater than the threshold , it is determined as "unknown fault", and the corresponding deviation vector is sent to the unknown fault candidate pool.

7. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 2, characterized in that, In the S11, the data acquisition adopts an edge computing gateway, the sampling frequency is set to 1 min / time, the data transmission delay is ≤500 ms, and the real-time operation data collected is subjected to outlier elimination, the elimination rule being that when a single data exceeds the mean of the same batch data ±3 times the standard deviation, it is determined as an outlier and is completed by linear interpolation.

8. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 2, characterized in that, In the S13, the mean calculation method of the operation data of the joint benchmark array is: ； ； ； wherein, n is the number of arrays participating in building the joint benchmark, , , Pi, Vi, Ii are the output power, operating voltage, operating current of the ith array, respectively, , is the variance of the performance ratio of the ith array, and the weight is used to highlight the proportion of array data with more stable performance.

9. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 5, characterized in that, The S3 further comprises: S34, knowledge base updating: collecting new fault samples every 3 months, updating the feature mean, covariance matrix and deviation feature rule in the fault knowledge base through incremental learning, and adding the new fault type label into the knowledge base; Incremental learning adopts a small sample learning model, and the updating is started when the number of new fault samples is ≥50 groups, the original fault feature data is retained during the updating process, and only the fault category parameters corresponding to the new samples are adjusted to ensure the compatibility of the knowledge base.

10. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 6, characterized in that, The unknown fault candidate pool is set with a regular review mechanism, the deviation vectors in the pool are subjected to cluster analysis every 1 month, if the number of vectors in a certain cluster is ≥30 groups, and the corresponding fault type is confirmed through artificial investigation, the vectors in the cluster are taken as new fault samples, and the knowledge base updating process of S34 is executed.

11. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 1, characterized in that, The method further comprises a fault early warning step S5: monitoring the change trend of the standardized deviation vector of the array to be diagnosed in real time, and triggering a fault early warning, outputting an early warning level and a pretreatment suggestion when the change trend of the standardized deviation vector of the array to be diagnosed in real time meets any one of the following conditions in a continuous 5 sampling periods a decrease amplitude ≥ 0.05 or an absolute value increase amplitude ≥ 5V. The warning level includes mild, moderate and severe.

12. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 11, characterized in that, In the S5, the warning level determination standard is: Light warning: the change range of deviation in 5 consecutive periods is 1-1.5 times of the warning threshold, and the current ; Moderate warning: the deviation change range is 1.5-2 times of the warning threshold in 5 consecutive periods, and the current ; Severe warning: the deviation change range is greater than or equal to 2 times of the warning threshold in 5 consecutive periods, and the current .

13. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 11, characterized in that, The dynamic benchmark array is re-screened once every 7 days, and if the performance of the original benchmark array is still within 0.98-1.02 and the variance is less than or equal to 0.015, the original benchmark array is retained as the benchmark array. If the performance of the original benchmark array decreases, the screening process of S12-S13 is re-executed to ensure the effectiveness of the benchmark array.

14. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 3, characterized in that, The calculation of the voltage deviation and the current deviation needs to be synchronized with the working period of the array to be diagnosed and the benchmark array, and only the data at the same time point with an irradiance of greater than or equal to 300 W / m2 and an ambient temperature of-10℃-45℃ are selected for comparison to exclude the interference of low irradiance and extreme temperature on the deviation calculation.

15. The photovoltaic array fault diagnostic method based on benchmark multi-dimensional deviation calculation according to claim 6, characterized in that, The S4 further comprises: S43, result output: output fault diagnosis result, including fault category, fault confidence , suggest troubleshooting site, and synchronize the diagnosis result to the photovoltaic power station operation and maintenance management platform; The fault confidence is divided into three levels: high confidence: confidence is greater than or equal to 0.8, medium confidence: 0.5 is less than or equal to 0.8, and low confidence: confidence is less than 0.5; when the confidence is low, secondary diagnosis is automatically started, 3 sets of real-time data are re-collected to calculate the standardized deviation vector, and the mean value of the 3 characteristic quantities is taken as the final judgment basis.

Images